Does Internal Migration Contribute to the Intergenerational Transmission of Socioeconomic Inequalities? The Role of Childhood Migration

Abstract

Introduction

With the gap between the highest and the lowest income groups widening in most OECD countries (Cohen and Ladaique 2018; Dorling 2015), the intergenerational transmission of advantage and disadvantage is a growing concern at the core of many economic and social policies (Western et al. 2007). The intergenerational transmission of socioeconomic inequalities is multidimensional, underpinned by such mechanisms as parental time and resource investments (Heckman and Mosso 2014), parental socialization (Coneus et al. 2012; Lareau 2003; Moullin et al. 2018), wealth transmission (Nolan et al. 2020), assortative mating (Eika et al. 2019; Frémeaux and Lefranc 2020), and neighborhood effects (Ryabov 2020). I propose internal migration as an additional mechanism contributing to the intergenerational transmission of socioeconomic inequalities.

Growing evidence suggests that internal migration behavior is intergenerationally transmitted (Bernard and Vidal 2020; Myers 1999b): individuals who migrated in childhood and adolescence report more internal migrations throughout adulthood. Further, internal migration in adulthood promotes upward occupational mobility and higher earnings (Fielding 1992). Thus, individuals exposed to internal migration early in life are likely to report better socioeconomic outcomes in adulthood. This untested hypothesis has important implications for understanding social mobility and social reproduction—that is, the reproduction of social inequalities across generations. On the one hand, internal migration could be an important mechanism for individuals to achieve intergenerational upward social mobility. On the other hand, because migration likelihood increases with educational level (Long 1973; Machin et al. 2012; Wagner 1990) and because this relationship is nearly universal (Bernard and Bell 2018), children from lower socioeconomic backgrounds are less likely to migrate than children from higher socioeconomic backgrounds. Thus, the potential for children to be doubly disadvantaged by having lower socioeconomic status (SES) and less exposure to internal migration might reinforce social reproduction.

I aim to test these hypotheses in the European context. Europe's heterogeneity in internal migration level makes it a particularly relevant case study. The spatial gradient of its internal migration level is marked, enduring, and characterized by higher levels in the northern and western countries of Europe and moderating levels in the southern and eastern countries (Bell et al. 2015; Bernard 2017; Rees and Kupiszewski 1999; Sánchez and Andrews 2011). In addition, Europe is unique in collecting high-quality migration data with broad geographic coverage (Bell et al. 2014; Poulain et al. 2013), which are essential for testing hypotheses on the mediating role of internal migration in the transmission of socioeconomic inequalities. Finally, the Nomenclature of Territorial Units for Statistics (NUTS) framework offers some degree of spatial and social heterogeneity between European regions, circumventing the problem of cross-national comparability of internal migration (Bell et al. 2002; Courgeau 1973; Courgeau et al. 2012). Thus, internal migration is defined in this paper as a change of NUTS-2 region of residence, the second tier of the NUTS framework.

I draw on the Survey of Health, Ageing and Retirement in Europe (SHARE), a series of nationally representative longitudinal surveys of the population aged 50 and older in 27 countries. Common interview practices, similar questionnaires, and standardization of fieldwork practices (Börsch-Supan et al. 2013; Börsch-Supan and Jürges 2005) have made SHARE the largest comparable longitudinal migration data set in Europe. In addition to its longitudinal modules, SHARE retrospectively collected in 2008/2009 (Wave 3) and 2017 (Wave 7) complete migration, employment, marital, and parental histories since birth, using life history grids to facilitate recall. SHARE also collects data on parental SES.

The analysis proceeds in two complementary steps to ensure the results are robust to model specification. First, I use an ordered logistic regression to assess the association between childhood migration and adulthood SES decile while controlling for parental background. Second, recognizing the interactions between explanatory variables, I use structural equation modeling to quantify how parental SES is transmitted via the mediating role of childhood migration and its impact on adulthood migration. This approach also permits controlling for other pathways through which parental SES affects later-life socioeconomic outcomes to ensure that the effect of childhood migration is not overestimated. I then use the model to explore variation in the frequency and direction of moves between rural and urban centers.

Conceptual Framework

In this section, I synthesize three lines of inquiry: (1) the intergenerational transmission of internal migration, (2) the link between spatial and social mobility, and (3) the selectivity of internal migration. I then leverage these strands of research to develop new hypotheses that link internal migration to social reproduction.

Intergenerational Transmission of Internal Migration Behavior

Parent-to-child transmission of demographic behaviors is well-established, particularly with respect to marriage (Bernardi 2016), fertility (Vidal et al. 2020), and divorce (Dronkers and Härkönen 2008; Lyngstad and Engelhardt 2009), but evidence for migration is more limited. According to socialization theory, children progressively internalize parental beliefs, values, and attitudes as normative behavior through repeated parent–child interactions, thereby coming to mirror parental behavior in adulthood. The well-established age schedule of migration (Rogers and Castro 1981) implicitly links parents to children who are dependent migrants. Further, emerging research has argued that by serving as behavioral models, parents transmit their views on migration, shaping their children's migratory behavior in adulthood. Despite my focus on internal migration in this article, this section draws on parallel literatures regarding residential mobility and international migration, where most progress has been made. Still, I recognize differences and similarities in the socialization into different forms of population movement.

Socialization into higher migration propensities results from two intertwined processes: parent-to-child transmission and children's lived experiences, which reinforce intergenerational transmission. De Jong (2000) observed that immediate family members' migration perceptions are strong predictors of internal migration in Thailand. Although direct measures of family migration norms are rare, studies have demonstrated parent–child similarities in migration behavior. For example, second-generation immigrants from majority ethnic groups in Kosovo and Latvia are more likely to emigrate than the native-born population (Ivlevs and King 2012, 2015). Likewise, in the United States, parents and children show similarities in residential mobility. Myers (1999b:871) concluded that residential mobility is “a lifestyle transmitted from parent to child, similar to how gender, political, and religious behaviors are transmitted.” Similarly, Ivlevs and King (2012:128) argued that the “intergenerational transmission of family migration capital emerges as a migration driver distinct from other known determinants of [international] migration . . . intrinsic to the migration process itself.” Because internal migration occupies an intermediary position between residential mobility and international migration, Bernard (2022b) argued that similar mechanisms are likely to operate on internal migration because of the acquisition of perceptions and skills that facilitate subsequent migration.

Children's lived experiences reinforce the parent-to-child transmission of internal migration. Recent evidence from Europe shows that individuals who changed NUTS-2 region of residence in childhood report more internal migrations by age 50; this relationship held after the researchers controlled for parental background (Bernard and Vidal 2020). Although the strength of this association is moderated by the national context, the distance moved, and the timing of childhood moves, it appears to hold in virtually all European countries. Two types of mechanisms linking childhood internal migration experiences to later-life migration behavior have been proposed. From a demographic perspective, individuals who migrated in childhood are more likely to have a first internal migration in adulthood and do so at a younger age, increasing their probability of a second internal migration (Bernard 2022b). From a psychological perspective, Myers (1999b) argued that residential mobility experiences alter the cost–benefit analysis regarding moves because past migrants have the knowledge and confidence to support relocation. By leaving and entering new social contexts, movers develop social skills that they can mobilize for future migration, reducing the migration-related stress and psychological consequences that are deterrents of residential mobility (Oishi 2010; Oishi et al. 2012). Because internal migration more extensively severs social ties than residential mobility, social skill acquisition through lived experiences is expected to be greater. This perspective aligns with the internal migration literature's perspective that migrants learn by doing (Bailey 1989; Bernard and Perales 2021; Morrison 1971).

This learning process is not restricted to childhood. Migration-facilitating skills and perceptions are acquired throughout life (Bailey 1989). As Morrison (1971:179) noted regarding internal migration, “decision thresholds are initially high for persons who never moved in their adult life. . . . Once a move has been made, though, the experience may foster a learning process that blunts subsequent inertia.” However, childhood and adulthood migration differ in two respects. First, as the first migration experience, childhood migration leaves a long-lasting impact. For example, preschool internal migration experiences affect adulthood internal migrating behavior in the Czech Republic, Denmark, France, Spain, and Switzerland (Bernard and Vidal 2020). Second, because it occurs in the family context, childhood migration facilitates the intergenerational passing of preferences, attitudes, and norms on ideal migratory behaviors. In other words, lived migration experiences in childhood reinforce socialization into migration behavior. Thus, internal migration experiences in childhood facilitate the acquisition of migration-facilitating skills and perceptions that can be mobilized later in life. Therefore, I expect the following:

• Hypothesis 1 (H1): Individuals who migrated in childhood are more likely to migrate in adulthood.

This process is represented by arrow 2 in Figure 1, which depicts the conceptual framework I propose.

However, the likelihood to migrate within a country increases with educational attainment (Greenwood 2016; Machin et al. 2012; Wagner 1990)—a nearly universal relationship (Bernard and Bell 2018; González-Leonardo et al. 2022). Thus, not all children are equally likely to be exposed to internal migration. On the basis of the positive relationship between education and internal migration, I posit the following:

• Hypothesis 2 (H2): Children from higher socioeconomic backgrounds are more likely to migrate than children from lower socioeconomic backgrounds.

This hypothesis is represented by arrow 1 (Figure 1). Note that unemployed individuals report a high propensity to move (Goss and Paul 1990; Hugo and Bell 1998). More generally, individuals from disadvantaged backgrounds display heightened mobility levels caused by job loss and insecure housing tenure. However, most of these moves are short-distance residential moves (Kull et al. 2016; Wiesel 2014) and are therefore not directly relevant to the links discussed here.

Childhood migration can also leave a lifelong imprint if it improves personal circumstances by relocating the child to a region with greater educational and employment opportunities. Where one grows up substantially influences subsequent life chances and outcomes (arrow 4 in Figure 1). Studies have found geographic variations in social mobility rates within Australia (Deutscher and Mazumder 2020), Canada (Corak 2020), Denmark (Eriksen and Munk 2020), England (Bell et al. 2022), and the United States (Chetty et al. 2014). In England, individuals born outside London and southeastern England who migrated out of their birth region report, on average, higher rates of upward social mobility than those who stayed (Buscha et al. 2021). Considering sociospatial inequalities and the greater educational and employment opportunities that cities provide, I predict the following:

• Hypothesis 3 (H3): Children born in or migrating to urban centers are less likely to migrate in adulthood than those born in or migrating to rural areas.

Conversely, by severing social ties and networks, childhood relocations can sometimes have a detrimental impact on school performance, mental health (Tseliou et al. 2016), and social integration later in life (DeWit 1998; Myers 1999a), although some studies have found no statistically significant relationship (Anderson et al. 2014; Gambaro and Joshi 2016; McMullin et al. 2021) or have even found a positive association (Swanson and Schneider 1999; Vidal and Baxter 2018). Although early studies focused on the weak but negative effect of internal migration, there is growing agreement that its impact is conditioned by the timing, context, and frequency of moves. Only moves that coincide with key developmental periods (Rumbold et al. 2012) appear to exert a negative influence. In addition, the impact of childhood migration is strongly modulated by the socioeconomic context. Residential relocations associated with changes in family structure, such as parental separation (Pribesh and Downey 1999; Tucker et al. 1998), insecure housing tenure (including evictions; Schwartz et al. 2022), and unemployment periods (Ziol-Guest and McKenna 2014) are associated with children's poorer life outcomes. Thus, the detrimental impact of residential mobility on children's life outcomes does not hold once family stressors and socioeconomic disadvantage are controlled for (Gambaro et al. 2022; McMullin et al. 2021; Vidal and Baxter 2018). More importantly, childhood movement linked to disadvantage appears to be more relevant to residential mobility than long-distance internal migration because individuals facing unemployment and insecure housing typically move within their local area (Kull et al. 2016; Wiesel 2014). However, the residential mobility literature has shown that repeated moves—eight by age 12 (Tucker et al. 1998)—can be detrimental to children (Mollborn et al. 2018; Pribesh and Downey 1999; Tseliou et al. 2016). Extending this argument to internal migration and considering the repeated severance of social ties, I posit the following:

• Hypothesis 4 (H4): The positive impact of childhood migrations on adult migration decreases with the number of childhood migrations.

Internal Migration and Social Mobility

The intergenerational transmission of migration behavior has important socioeconomic implications because internal migration in adulthood is associated with, on average, better life outcomes—in particular, upward occupational mobility (Mulder and van Ham 2005) and higher earnings (Greenwood 2016). This association has been explained by migration to so-called escalator regions that provide ample educational and employment opportunities, such as southeastern England (Fielding 1992) and northern Italy (Impicciatore and Panichella 2019), and play a significant role in improving labor market outcomes over individuals' life course (Van Ham et al. 2012). When employment opportunities are unevenly distributed across regions, migration plays a particularly important role by allowing internal migrants to extend their job search geographically (Ballarino and Panichella 2021; van Ham 2003). Economic studies controlling for endogeneity issues with instrumental variables have shown that the positive impact of migration on labor market outcomes holds even after controlling for selectivity (Borjas 1987; Borjas et al. 1992). Similarly, using a quasi-experimental design, Panichella and Cantalini (2022) have shown that the positive association between geographic and social mobility holds for Italian males of different social origins. Thus, I posit the following (represented by arrow 5 in Figure 1):

• Hypothesis 5 (H5): Individuals who migrate internally in adulthood report higher socioeconomic outcomes than stayers.

The strength of this association is likely to be moderated by the direction of migration. Drawing on the escalator region framework (Fielding 1992), I anticipate that urban centers provide access to more and better educational and employment opportunities and therefore predict the following:

• Hypothesis 6 (H6): Migrating to urban centers exerts a stronger positive effect on SES than migrating to rural areas.

If the empirical analysis supports H1 (internal migration is intergenerationally transmitted) and H5 (internal migration in adulthood improves labor market outcomes), then I will be able to conclude that individuals exposed to early-life internal migration report better economic outcomes in adulthood than individuals who were stayers in childhood. If the results also confirm that children from higher socioeconomic backgrounds are more likely to migrate in childhood (H2), then I will be able to conclude that children from lower socioeconomic backgrounds are doubly disadvantaged in having a disadvantaged background and being less migratory, which exacerbates social reproduction by limiting their mobility in adulthood and the economic benefits of changing region of residence.

Data and Methods

SHARE Data

I draw on data from SHARE, a suite of surveys that are nationally representative of the population aged 50 years and older in each country. Wave 7 of SHARE retrospectively collected the complete life histories of respondents in 26 European countries¹ who were born before 1968. All respondents who did not participate in the Wave 3 retrospective module received the module in Wave 7. These individuals include respondents from 15 countries that did not participate in Wave 3 and joined SHARE between Waves 4 and 7, as well as respondents from refreshment samples drawn between Waves 4 and 6 from another 12 countries. Thus, although SHARE is a longitudinal survey, I draw on only one wave of retrospective data, and the analysis is unaffected by between-wave attrition. I therefore use cross-sectional weights for descriptive statistics to match the size of national target populations aged 50 and older in 2017 (Bergmann et al. 2019). As is common in life course studies (Lynch 2003; Rosenfeld and Roesler 2019) and as recommended by Winship and Radbill (1994) when sampling weights are a function of independent variables, regression analyses are unweighted.

Nine countries (Hungary, Lithuania, Bulgaria, Cyprus, Finland, Latvia, Malta, Romania, and Slovakia) did not collect parental education, and two (Poland and Portugal) collected patchy parental education records. Thus, the analysis excludes these 11 countries and is restricted to 15 countries: Austria, Belgium, Croatia, the Czech Republic, Denmark, Estonia, France, Germany, Greece, Italy, Luxembourg, Slovenia, Spain, Sweden, and Switzerland. To obtain life histories of comparable length for all respondents, I restricted the analysis of life histories from birth to age 50. To avoid truncation, I restricted the analysis to individuals born in the survey countries. The final analytic sample comprises 23,145 individuals. Although all retrospective data are limited by missing data on individuals who died or moved beyond the survey catchment geography (in this case, outside the 15 case study countries), SHARE provides invaluable information by collecting complete migration histories beginning at birth and parental SES.

Wave 7 retrospectively collected life histories encompassing (1) housing and migration, (2) education and employment, (3) partners and children, and (4) health and health care. Data were collected using life history grids, depicting survey responses schematically from birth to present to help respondents recall past moves (Belli 1998; Blane 1996). Rows in the grids represent the different life domains against which retrospective data are collected, and columns represent successive years of life. This sequential, cross-dimensional visual representation, which displays both migration events and spell (i.e., duration of residence), has been shown to reduce recall error and improve data quality by allowing respondents to check dating accuracy relative to other events that act as temporal landmarks (Axinn et al. 1999; Brüderl et al. 2017; Glasner et al. 2012; Shum 1998). Comparison of contemporaneous information collected in Waves 1 or 2 with retrospective information collected in Wave 3 for the same individuals suggests high recall quality in SHARE (Börsch-Supan 2020). However, recall accuracy is higher for more recent moves than for those in the distant past (Smith and Thomas 2003), and as similarly found for recall accuracy in fertility studies (Schoumaker 2014), omission of moves to reduce survey burden may cause the underestimation of migrations. However, recent validation studies comparing the incidence of life course events between retrospective and longitudinal data sources have shown the potential for accurate retrospective collection of salient key life course events (Börsch-Supan 2020), including childhood events (Smith 2009), such as long-distance migration.

Measures

Recognizing that SES is a multifaceted concept related to one's access to cultural, financial, and social resources, I construct a composite childhood and adulthood SES measure. This approach better reflects the multidimensional nature of socioeconomic advantage and disadvantage than single variables (Cowan et al. 2012). Mazzonna (2014) was the first to apply this approach to SHARE data, and others subsequently validated (Havari and Mazzonna 2015) and implemented it (Angelini et al. 2019). For childhood, I conduct a principal component analysis on four indicators measured at age 10: the number of rooms, the number of facilities (fixed bath, cold and hot running water supply, inside toilet, and central heating), the number of books in the home, and the main breadwinner's occupation. Parental education at birth is reported separately, as explained later. I similarly obtain SES at age 50 by running a principal component analysis on the main occupation, the highest level of education attainment, and homeownership.

Parents' and respondents' occupations were categorized using the ISCO-08 classification (International Labor Organization 2007), which is used to calculate the Standard International Occupational Prestige Score (Treiman 1977). SIOPS is a hierarchical scale ranging from 0 to 100, with higher values indicating higher occupational prestige. This index is stable across cohorts (Lersch et al. 2020), countries (Hout 2003), and gender (Warren et al. 1998). I exclude household income because it was not collected retrospectively (Härkönen et al. 2016). Education was collected using the ISCED-97 classification (up to primary, secondary, postsecondary nontertiary, and tertiary). Following the Kaiser rule, I retain only factors with an eigenvalue greater than 1, which corresponds to one factor for childhood SES and one factor for SES at age 50. Factor loadings for these indicators are shown in online appendix A. For ease of interpretation, I categorize the SES indices into deciles, with 1 indicating the most disadvantaged decile and 10 indicating the most advantaged decile. This approach has advantages over a continuous variable in permitting the effective visualization of results and establishing whether the association between childhood and adulthood SES is linear.

I define internal migration as a change in the NUTS-2 region of residence within Europe. SHARE collected up to 30 changes of address for durations of residence of six or more months beginning at birth, recording the NUTS-2 region of residence for each change of address. The NUTS framework offers spatial and social heterogeneity between regions. This spatial framework circumvents the problem of comparability of internal migration between countries (Bell et al. 2002; Courgeau 1973; Courgeau et al. 2012) and captures internal migration, which differs from residential mobility in being a long-distance move that severs social ties. NUTS-2 regions range from 800,000 to 3 million inhabitants. These regions generally mirror the territorial administrative division of the European Union member states and correspond to regions in France, Hungary, and Italy; provinces in Belgium; national areas in Sweden; and autonomous communities in Spain. NUTS-2 is the most commonly used administrative unit for studying internal migration in Europe, particularly in a cross-national framework (Van der Gaag and van Wissen 2008).

In the baseline model, I measure migration histories as binary variables: whether the individual changed NUTS-2 of residence at least once. Because SES is measured at age 10, I distinguish between migration history at birth–age 9 and at ages 10–17. Migration history in adulthood is measured at ages 18–50. In a subsequent model, I enter the number of migrations as binary variables to detect possible nonlinearity. Finally, I consider migration direction by distinguishing between moves to and from urban and rural areas by separately exploring urban-to-urban, rural-to-urban, urban-to-rural, and rural-to-rural migration based on self-reported urban status.² Descriptive statistics in Table 1 show that internal migration is a rare event: slightly more than 10% of Europeans migrated in childhood, less than 25% migrated in adulthood, and 12% engaged in repeat migration in adulthood. The direction of movement is broadly spread, although rural-to-rural migration is, unsurprisingly, less common.

Individual-level control variables include sex, birth cohort, parental status (i.e., ever had a child) at age 50, and marital status (i.e., ever married) at age 50. To adjust for possible confounders of the relationship between adulthood internal migration and social position, I also control for personality traits, which influence migration (Bernard 2022a; Campbell 2019; Jokela 2021; Shuttleworth et al. 2020) and job mobility (Huinink et al. 2014). Wave 7 was the first SHARE wave to assess personality traits using the 10-item Big Five Inventory (BFI-10) (Rammstedt and John 2007). Personality traits are relatively stable over time (Cobb-Clark and Schurer 2012) and are not influenced by migration experiences (Crown et al. 2020). See online appendix B for descriptive statistics for all explanatory variables.

Analytic Strategy

where $i$ corresponds to individuals, $X$ is a set of binary variables capturing childhood SES decile at age 10, $V$ is a vector of individual-level control variables (sex and birth cohort), and $P$ is a vector of personality traits. $β,δ$⁠, and $ρ$ are the corresponding vectors of regression coefficients. The country fixed effect is captured via country dummy variables in vector C with a corresponding vector of regression coefficients $σ$⁠. Finally, α is the model's intercept, and e is the error term.

Second, to further explore the mediating role of childhood migration histories in social reproduction, I use structural equation modeling to conduct mediation analysis. This approach uses a system of interrelated equations to estimate (1) the indirect effect of childhood SES on adulthood SES via childhood migration histories and the region type where children grew up and (2) the effect of childhood migration history on adult migration behavior, while controlling for the direct effect of childhood SES on adulthood SES. Upcoming Figure 3 offers a visual representation of this multiequation system, with regression coefficients of interest displayed along the arrows. As explained earlier, because SES is recorded at age 10, the role of childhood migration histories is modeled separately before and after age 10. Therefore, parental education is a proxy for SES at birth.

This framework is estimated as a system of seven interrelated equations: Eqs. (2) to (8). I estimate binary dependent variables—migration histories and urban status—via binary logistic regression and estimate SES decile via ordered logistic regression in Eq. (2). All regressions control for sex and birth cohort in vector $V$ and include a country fixed effect captured in vector C. For outcomes measured in adulthood (SES and migration history), marital and parental histories are controlled for in vector H and personality traits are controlled for in vector $P$⁠. All equations include an error term, $e$⁠.

As shown in Eq. (2), adulthood SES decile, $Y$⁠, is an ordinal variable with $c$ categories that represents the 10 SES deciles and $j=1,2, . . . ,c−1$⁠. It is expressed as a function of childhood SES decile at age 10 through vector $X$⁠, adult migration history through variable $M$⁠, and a corresponding regression coefficient $μ$⁠. Urban status of the region of residence at age 18 is represented by T and the corresponding regression coefficient $π1.$$H$ is a vector of marital and parental histories. The regression coefficient $μ1$ quantifies the impact of adult migration on SES to test H5.

Migration history in adulthood,$M$⁠, is expressed in Eq. (3) as a function of migration history at ages 10–17 and at birth–age 9, which are captured, respectively, by $Z$ and $Q$ and corresponding regression coefficients $ϑ1$and $τ1$⁠. I use these regression coefficients to test H1, which suggests that individuals who migrated in childhood are more likely to migrate in adulthood. In Eq. (3), I control for urban status at age 18, $T$⁠. The corresponding regression coefficient$π2$ permits testing H3 by revealing whether children born in or migrating to urban centers (relative to rural areas) are less likely to migrate in adulthood.

Urban status at age 18 is a product of migration history at ages 10–17:

Migration history at ages 10–17 (⁠$Z$⁠) is a function of SES at age 10 (⁠$X$⁠) and region of residence at age 10 (⁠$W$⁠), as shown in Eq. (5), with corresponding regression coefficients $β3$ and $ω1$⁠. Testing H2, the vector of regression coefficients $β3$ captures the effect of parental SES on internal migration exposure:

As shown in Eq. (6), SES at age 10 $(X)$ is a function of the vector of parental education at birth $(S)$ and urban status of region of residence at age 10. The latter is a function of migration history from birth to age 9 $(Q)$⁠, as shown in Eq. (7):

As shown in Eq. (8), migration history from birth to age 9 (⁠$Q$⁠) is a function of the urban status of the region of birth $(K)$ and the vector of parental education $(S)$⁠, with corresponding regression coefficients $γ$ and $φ2$⁠. The vector of coefficients $φ2$ tests H2.

In subsequent model specifications, I replace the binary variable capturing migration histories with a series of binary variables from zero to four or more migrations in Eqs. (2) and (7). These models test H4, which suggests that the positive impact of childhood migration experiences on adult migration decreases with the number of childhood migrations. Finally, I consider the direction of migration, distinguishing between moves to and from urban and rural areas in Eqs. (2) and (7). I thereby establish whether migration to urban centers exerts a stronger positive effect on socioeconomic outcomes than migration to rural areas, as H6 suggests.

To facilitate interpretation, I exponentiate regression coefficients to produce odds ratios (ORs). ORs larger (smaller) than 1 indicate that a given explanatory variable is associated with an increased (decreased) likelihood of an outcome relative to the baseline outcome.

Empirical Results

Descriptive Statistics

To help interpret results from the regression analysis, Figure 2 presents descriptive statistics that tease out the conceptual framework's proposed links. Panel a displays the childhood SES decile (where 1 is the most disadvantaged decile and 10 is the most advantaged decile) by the average number of childhood internal migrations. The number of childhood internal migrations increases with SES. Children from the top two deciles, on average, report roughly twice as many internal migrations as children from the lower deciles, suggesting that the propensity to migrate internally in childhood is socially determined. Panel b displays a clear linear association between the number of childhood migrations and the number of adulthood migrations. Individuals who did not migrate in childhood report an average of 0.40 migrations in adulthood, compared with one or more migrations among individuals who migrated at least twice in childhood. This finding supports the idea that internal migration is intergenerationally transmitted. Finally, panel c depicts a positive linear association between the number of adulthood internal migrations and SES at age 50: individuals who migrate in adulthood multiple times are more likely to report higher SES at age 50. The next step is to explore these associations in a regression framework and control for possible confounders.

Regression Analysis

Table 2 reports regression coefficients from ordered logistic regression models. Results from Model A1, which controls only for sex and birth cohort, reveal a positive association between SES at age 50 and childhood internal migration (OR = 1.12, p < .05). The result holds when I control for childhood SES in Model A2 and personality traits in Model A3. As expected, SES at age 50 increases with SES at age 10 and openness to new experiences. Each additional SES decile increases adulthood SES, but the effect is particularly pronounced among the top two deciles. Finally, the addition of a country fixed effect in Model A4 does not change the results. The inclusion of these additional controls improves the model fit, as indicated by a decrease in the Akaike information criterion and the Bayesian information criterion.

Baseline Model

To identify the mechanisms through which childhood migration histories influence SES, Figure 3 reports regression coefficients from the structural equation modeling outlined in Eqs. (2)–(8). (See online appendix C for full regression results.) Control variables have the expected sign. In particular, the higher a child's SES, the higher their SES at age 50.

Results in Figure 3 confirm that children from higher SES backgrounds are more likely to be exposed to internal migration than children from lower SES backgrounds: the higher the level of parental education, the greater the odds of migrating at birth–age 9. The effect of education is gradual. The likelihood of migrating before age 10 is 1.60 times as likely for those with secondary-educated fathers (OR = 1.60, p < .001) and 2.6 times as likely for those with tertiary-educated fathers (OR = 2.62, p < .001) relative to children whose fathers completed primary education (only the latter OR is shown in Figure 3; see legend details). Results are similar for maternal education. The association between SES at age 10 and migration at ages 10–17 is also positive and statistically significant (OR = 1.04, p < .001). These results support H2: children from higher SES backgrounds are more likely to migrate than children from lower SES backgrounds.

Children exposed to internal migration are more likely to migrate internally in adulthood, confirming the migration-as-a-learned-behavior proposition (H1). Migration in early and late childhood is significantly associated with greater odds of adulthood migration. Individuals exposed to internal migration before age 10 are 52% more likely to migrate in adulthood (OR = 1.52, p < .001); individuals who migrated at ages 10–17 are twice as likely to migrate in adulthood (OR = 2.23, p < .001). The greater effect of migration in late childhood suggests that more recent migrations bear more weight on subsequent migration decisions. Another possibility is that older children and adolescents are more aware of the challenges and benefits of migration than children who migrated before age 10.

In addition, adulthood internal migration increases the odds of belonging to higher SES deciles at age 50 (OR = 1.46, p < .001), supporting H5. To further explore the mediating role of migration in social reproduction, I replace SES decile membership with tertiary educational attainment, homeownership, and occupational prestige. I model these outcomes separately in three distinct models and present the full results in online appendix D. I model the occupational prestige score using an OLS regression and binary outcomes (homeownership and tertiary education) using binary logistic regression. The positive and statistically significant coefficients for tertiary education (OR = 1.85, p < .001), homeownership (OR = 1.13, p < .001), and occupational prestige (β = 1.73, p < .001) indicate that adulthood internal migration contributes to all dimensions of social standing.

The results of these analyses confirm the mediating role of internal migration in social reproduction. Individuals exposed to internal migration in childhood report, on average, better socioeconomic outcomes in adulthood than individuals who were not.

Number of Migrations

To establish whether the observed association holds for repeat migrants, I run an additional model that replaces migration histories with binary variables capturing the number of migrations in childhood and adulthood (none, one, two, three, or four or more). This tests H4, which proposes that the positive impact of childhood migration experiences on adult migration decreases with the number of childhood migrations, given the repeated severance of social ties. Because SES is captured at age 10, the baseline model distinguishes migration before and after age 10. However, few children migrate multiple times in each period, and rare events can produce biased estimates because of the lack of statistical power and overfitting (King and Zeng 2001; O'Brien 1986). To circumvent this issue, I group childhood migration from birth to age 17. Given that migration experiences from birth to age 9 likely influence SES at age 10, I instead measure SES as paternal and maternal education at the child's birth. Although it is a cruder measure than the composite index at age 10 based on home facilities, the number of rooms and books, and the breadwinner's main occupation, sociological and demographic studies commonly use it to capture parental SES (Aschaffenburg and Maas 1997). Online appendix E presents a visual depiction of this reduced model.

Key regression coefficients of interest are reported in Figure 4. (See online appendix F for full results.) Panel a shows that the positive association between adulthood migration and SES at age 50 holds for moves of all orders. More importantly, migrating at least twice in adulthood has a significantly greater effect on SES than migrating once. However, confidence intervals show that migrating more than two times does not lead to significantly higher benefits. Panel b of Figure 4 shows that migrating at least once in childhood is paramount to increasing the odds of migrating in adulthood. However, the overlapping confidence intervals imply that the number of childhood migrations is not relevant to subsequent migration decisions. Still, migrating multiple times in childhood is not detrimental despite the repeated severance of social ties, invalidating H4.

Direction of Migration

Of particular interest in Figure 3 is the positive and statistically significant association between urban residence at age 18 and SES at age 50 (OR = 1.37, p < .001). Results in online appendix D show that this positive relationship is negative for homeownership, likely because of higher prices in urban centers. However, the positive relationship holds for occupational prestige and tertiary education completion, highlighting the importance of access to educational and employment opportunities early in adulthood for shaping later-life socioeconomic outcomes.

Results also indicate that urban residence in childhood is positively associated with SES at age 10 (OR = 1.32, p < .001). A possible explanation is that children who grew up in urban centers, which offer greater educational opportunities for children and more employment opportunities for their parents, are more advantaged while growing up, although this conjecture cannot be formally tested here. Not surprisingly, children who migrated internally from birth to age 9 are much more likely to reside in an urban region at age 10 (OR = 1.70, p < .001). Thus, children who grew up in urban centers do not need to migrate to experience upward social mobility, consistent with H3. However, residence in an urban center at age 18 does not decrease the odds of adulthood migration: young adults residing in urban centers are almost 15% more likely to migrate throughout adulthood than their rural counterparts (OR = 1.15, p < .001), reinforcing the influence of past migration on future migration decisions (H1).

I further explore sociospatial inequalities by replacing migration history (having migrated or not) with the direction of moves. (See online appendix G for the full results.) Selected results in Figure 5 show adulthood migration in any direction is positively and statistically significantly associated with SES at age 50. As might be expected, the effect is significantly lower for rural-to-rural migration (OR = 1.15, p < .001) than for other flows. However, confidence intervals indicate that urban-to-urban (OR = 1.61, p < .001), rural-to-urban (OR = 1.69, p < .001), and urban-to-rural (OR = 1.53, p < .001) migration do not significantly differ. Wald tests (in online appendix H) confirm these results. Unsurprisingly, migration between urban regions and migration from rural to urban regions are equally beneficial to SES: both types of movement provide access to the educational and employment opportunities that urban centers offer. A perhaps less expected result is the strength of the effect of urban-to-rural migration. Although rural stayers typically earn less than rural-to-urban migrants (Culliney 2017), the results may indicate that some individuals leave once they have little more to gain from remaining in urban areas or when liquidating assets toward the end of their working life (Champion 2012; Culliney 2017; Fielding 1992). However, because the analysis ends before retirement (at age 50), the association observed between urban-to-rural migration and SES may also be linked to the counter-urbanization (Champion 2001) that has unfolded in western and southern Europe (Rowe et al. 2019); these processes cannot be formally tested here. Overall, these results only partially align with H6.

Conclusion and Discussion

This article combines three lines of inquiry—(1) the intergenerational transmission of internal migration behavior, (2) the link between spatial and social mobility, and (3) the selectivity of internal migration—to propose that internal migration mediates the intergenerational transmission of social inequalities. Drawing on retrospective life history data from 15 European countries, this study provides empirical evidence supporting this proposition, showing that children from higher socioeconomic backgrounds are more likely to migrate internally, increasing their chances of adulthood migration, which is associated with higher SES by age 50. What is essential in setting individuals on a future migration trajectory is exposure to the benefits and challenges of internal migration at least once in childhood, more so than the number of early-life internal migrations. Table 3 summarizes these findings.

Thus, internal migration behavior is, to some extent, socially determined. Freedom of movement, recognized by the 1948 Declaration of Human Rights, is constitutionally entrenched in many democratic countries and supported by laws and regulations. Yet, this study reveals the social barriers to migrating internally—a dimension rarely considered in the internal migration literature. In particular, this work shows that childhood internal migration is rare, with just over 10% of Europeans having changed NUTS-2 region of residence before age 18 and most of them drawn from the top two SES deciles (see Figure 3).

Despite its mediating role, internal migration should not be considered the main driving force behind social reproduction but rather an additional factor that reinforces the transmission of advantage and disadvantage. Therefore, when possible, social mobility studies should control for migration experience, an additional source of advantage for children from privileged backgrounds. The cornerstone of this argument is the role of intergenerational transmission of internal migration—through parent-to-child transmission and a child's lived experiences—in facilitating the acquisition of migration-facilitating skills and perceptions that can be mobilized later in life. Despite growing interest in parent-to-child transmission of demographic behaviors, direct measures of family migration norms are rare. Further empirical research is required to confirm the channels through which childhood migration experience alters the cost–benefit analysis that accompanies later-life migration decisions.

This study's results show that the strength of the mediating role of internal migration is moderated by the direction of moves. Unsurprisingly, rural-to-urban and urban-to-urban migration returns a stronger association with SES at age 50 than rural-to-rural migration, presumably because of urban centers' greater educational and occupational opportunities. This result is an important reminder that economic opportunities are spatially embedded. As a response to sociospatial inequalities, internal migration can thus further exacerbate social reproduction. The results show that children who migrated are more likely to reside in an urban center at age 18, which is associated with higher SES later in life. Thus, childhood migration can also increase inequalities without adulthood migration. From a policy perspective, these findings reinforce the importance of place-based interventions that target children from low-SES backgrounds living in the most disadvantaged regions. Such policy programs are increasingly important because of the continued widening of polarization between metropolitan and nonmetropolitan regions in most European countries (Kühn 2015; Martin et al. 2022).

Because the mediating role of internal migration in social reproduction is likely modulated by the macro context, I use a country fixed effect. Future work should further explore the role of social and economic policies. Countries with sociodemocratic welfare regimes, such as Sweden and Norway, intervene to equalize opportunities and address inequalities. In such countries, the role of internal migration in social reproduction may be weaker than in liberal welfare regimes found in Anglo-Saxon countries, which report lower levels of social mobility (Esping-Andersen 2015; Western and Wright 1994). Yet, SHARE does not include countries with a liberal welfare regime. Thus, an important direction for future research is to replicate the analysis presented here in Anglo-Saxon countries, where the incidence of childhood poverty is typically higher (Fouarge and Layte 2005). This line of inquiry would benefit from an increased availability of retrospective surveys. Such surveys include the English Longitudinal Study of Ageing and its sister surveys: the Life Histories and Health survey in Australia and the Life History Mail Survey collected in the United States as part of the Health and Retirement Study.

Another promising data set is the German Life History Study, which includes cohorts born in 1919–1971 (Mayer 2015). Use of this data set should permit exploring intercohort change in the mediating role of internal migration and focusing on more recent birth cohorts. The current study's results pertain mainly to baby boomers. Examining the replicability of these findings in various historical, geographic, and institutional contexts is an important direction for future research, which is particularly relevant considering the increase in the selectivity of internal migration observed in some countries.

For example, the long-term decline in internal migration in the United States (Champion et al. 2018; Cooke 2011) has been pronounced among part-time, low-income, and low-skilled workers (Foster 2017)—a pattern explained by stricter occupational licensing regulations (Schleicher 2017) and housing price increases in productive regions (Ganong and Shoag 2017). Thus, U.S. children from lower SES backgrounds are increasingly less likely to migrate internally. This shift is likely to reduce their adulthood migration, thereby reducing their chances of social mobility proportionally more than children from higher SES backgrounds, who remain comparatively more migratory. As a result of this process, internal migration might contribute to the widening of socioeconomic inequalities. However, in Australia and the United Kingdom, the decline in internal migration has not been accompanied by a withdrawal of lower SES groups from the internal migration system (Kalemba et al. 2020; Shuttleworth et al. 2019). Drawing firm conclusions is complicated by the lack of research on trends in the selectivity of internal migration. Early theoretical expectations were that the educational selectivity of internal migration would decline as countries developed (Gould 1982; Long 1973). However, recent cross-sectional evidence of countries at varying levels of development suggests that the educational selectivity of internal migration does not abate as education expands (Bernard and Bell 2018) but follows a J-shaped relationship (Abel and Muttarak 2017). That is, the difference in the odds of migrating between individuals with low versus high educational attainment may be the highest in highly developed countries. Although further research is required to substantiate long-term trends in the selectivity of migration, these studies suggest that internal migration may contribute to the widening of socioeconomic inequalities by reducing low-SES children's exposure to internal migration.

Another important angle for future research to consider is gender. The positive effect of internal migration found in this study may be moderated by gender. Female internal migrants, particularly mothers, tend to benefit less in terms of earnings and occupational mobility than their male counterparts (Boyle et al. 2001), and in some instances, they suffer penalties in the labor market compared with nonmobile females (Cooke 2008). As tied migrants who follow a male breadwinner, females often seek new employment after migration, which tends to be detrimental to their earnings and occupational standing even if those losses are compensated by net gains in total family earnings (Panichella and Cantalini 2022). Even with more women in the labor force, men's labor market position and resources remain stronger determinants of family migration decisions than females' (Cooke 2008). Gender asymmetries in the role of spousal resources on family migration decisions are channeled by structural inequalities in the labor market (Nisic 2009; Perales and Vidal 2013), institutional environments (Cooke et al. 2009; Vidal et al. 2017), and gender ideology (Lersch 2016). Thus, future research should investigate the potential moderating effect of gender on the association between internal migration and social reproduction.

Another avenue for future research is to explore the relevance of the processes described here for international migration. Indeed, growing evidence suggests that international migration is transmitted intergenerationally (Ivlevs and King 2012), including for second-generation migrants with European parents (Bernard and Perales 2021; de Jong and de Valk forthcoming). Continuing increases in cross-country migration within Europe (DeWaard et al. 2012; DeWaard and Raymer 2012) point to the importance of understanding its potential role in the transmission of inequalities. Despite this study's focus on the native-born population, I have argued for a broader examination of migration as a socially determined endeavor contributing to social reproduction, with the aim of informing and spurring further research in the field.

Acknowledgments

The author thanks Professor Martin Bell and five anonymous reviewers for their comments and suggestions. This study uses data from SHARE Wave 7 (DOI: 10.6103/SHARE.w7.800); see Börsch-Supan et al. (2013) for methodological details. The SHARE data collection has been funded by the European Commission, DG RTD, through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812), FP7 (SHARE-PREP: GA N°211909, SHARE-LEAP: GA N°227822, SHARE M4: GA N°261982, DASISH: GA N°283646), and Horizon 2020 (SHARE-DEV3: GA N°676536, SHARE-COHESION: GA N°870628, SERISS: GA N°654221, SSHOC: GA N°823782, SHARE-COVID19: GA N°101015924) and by DG Employment, Social Affairs & Inclusion, through VS 2015/0195, VS 2016/0135, VS 2018/0285, VS 2019/0332, and VS 2020/0313. Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S. National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1-AG-4553-01, IAG BSR06-11, OGHA 04-064, HHSN271201300071C, RAG052527A), and various national funding sources is gratefully acknowledged (see www.share-project.org).

Notes

References