Intergenerational mobility is measured by the association between parents’ and adult children’s economic status, where status can be measured by income, earnings, occupation, class, education, wealth, or other markers of socioeconomic advantage. Mobility is an important concept because it provides information about equality of opportunity in society. A low intergenerational association suggests that individuals have an equal chance of succeeding (or failing) regardless of social origins. In contrast, a strong intergenerational association indicates that children will closely replicate their parents’ position in the socioeconomic hierarchy, suggesting the persistent influence of the advantages of birth. The recent increase in economic inequality in the United States and other advanced industrial countries [OECD 2008; McCall and Percheski Reference McCall and Percheski2010] underscores the importance of the question of mobility, based on the assumption that growing inequality will strengthen the intergenerational persistence of advantage. More generally, using mobility as a metric, the relationship between equality of outcomes and opportunity can be examined.
As is the case for inequality, mobility is a population-level comparative concept. While there will always be rags-to-riches and riches-to-rags biographies in societies, mobility measures, at the aggregate level, show how prevalent these stories are. Mobility is comparative because no empirical society is characterized by either null association or perfect association between parents and children. It is the relative comparison across countries or periods that provides bounds to gauge the level of intergenerational persistence in a particular context.
Scholars focus on “relative mobility,” i.e. the association or persistence between parents’ and adult children’s economic wellbeing net of any changes in the economic structure and in the aggregate levels of economic wellbeing across generations (sociologists call this component “structural mobility”). If rapid economic growth lifts all boats or if industrialization pushes individuals from agriculture to manufacturing and service employment, these components of mobility will be controlled away. This means that relative mobility cannot be directional. Each upward move requires a downward move that “frees up” a coveted space in the socioeconomic hierarchy, hence the metaphor of “taking turns” across generations to describe mobility [Hout Reference Hout and Neckerman2004]. This notion of intergenerational association has no correlate in the population’s experiences: mobility as a lived experience—and likely as a source of satisfaction, value orientations, and attitudes—includes both changes in aggregate standard of living and economic structure and net departures from social origins. Nor does it have a direct correlate in discussions about promoting upward mobility through, for example, educational expansion, because changes in educational distributions do not necessarily detach individuals from their social origins.
The analysis of mobility is largely bivariate, focusing on the description of the association between parents and children. This is no small feat: as I will discuss, producing a valid and comparable measure of intergenerational association across time and place carries substantial methodological and data challenges. A central methodological challenge is to devise a valid measure for parents’ and children’s long-term economic status (called “permanent” or “lifetime” income by economists), purged of errors of measurement and transitory fluctuations; this is the theoretical concept believed to shape children’s outcomesFootnote 1.
The analysis of mobility poses data challenges because it requires information on the economic circumstances of adult children and their parents when the children were growing up. Three data sources are used: long-term panel surveys following families over several decades; administrative data matching parents and children (for example, tax records); and cross-sectional surveys of adult children collecting retrospective information about their parents. The latter data source is the most common but the most error-prone. It permits the analysis of class and occupational persistence because children’s retrospective reports of parental occupation have acceptable validity and reliability. However, it does not allow for the analysis of income or earnings mobility. As a result, analysis of economic mobility is a “rich country” endeavor, largely restricted to about two dozen affluent countries (see for instance Blanden Reference Blanden2013, Bjorklund and Jantti Reference Bjorklund, Jantti, Salverda, Nolan and Smeeding2009, Black and Devereux Reference Black, Devereux, Card and Ashenfelter2011, Fox et al. Reference Fox, Torche, Waldfogel, Brady and Burtonforthcoming), and only recently expanding to middle-income countries such as China, Brazil, and Chile [Gong et al. Reference Gong, Leigh and Meng2012; Torche Reference Torche2014] by way of methodological innovations that link independent samples of parents and children [Angrist and Krueger Reference Angrist and Krueger1992].
Occupational and class mobility analysis can rely on cross-sectional surveys with retrospective information about parents, and so they have been conducted in many more countries, both developed and developing [Hout and DiPrete Reference Hout and DiPrete2006, Torche Reference Torche2014]. However, occupational and economic mobility are different concepts, and they are empirically uncorrelated across countries. As striking examples, the United States and Chile feature limited earnings mobility but higher levels of class mobility [Bjorklund and Jantti Reference Bjorklund and Jantti2000; Blanden et al. 2013; Torche 2005; Nunez and Miranda Reference Nunez and Miranda2010]. These discrepancies raise an important question concerning the validity and value of different measures of well-being in capturing the persistence of advantage across generations.
This article begins by reviewing economic and sociological perspectives for the analysis of intergenerational persistence and offering its main empirical findings. It then discusses topics that push mobility analysis beyond its bivariate focus and evaluate the utility of the concept: the relationship between cross-sectional inequality and intergenerational mobility, the mechanisms of mobility, the analysis of mobility across multiple generations, and the extent to which mobility captures equality of opportunity. The conclusions suggest that mobility is a useful descriptive measure of inequality of opportunity in society, which would ideally be available for every country in the world. However, it tells us very little about causal processes and mechanisms for the persistence of advantage. Addressing mobility mechanisms requires both isolating the causal effect of specific factors such as educational policy and family structure, and moving beyond these specific factors to assess how institutional contexts shape intergenerational opportunity.
Measurement and main findings of mobility research: sociological and economic perspectives
Sociology and economics have made important contributions to the analysis of mobility. Both disciplines focus on the association between parents and adult children, but they favor different measures of economic status and different strategies to gauge the intergenerational association. Sociologists have traditionally focused on occupational status and social class while economists focus on pecuniary measures such as wages, earnings and income, although these disciplinary boundaries are less rigid in current scholarship (e.g. Mitnik et al. Reference Mitnik, Bryant, Weber and Grusky2015; Bjorklund and Jantti Reference Bjorklund and Jantti2000).
Occupational status and class mobility
The concept of occupational status was originally devised by Duncan [Reference Duncan and Reiss1961] in the context of the American status attainment tradition. Occupational status is a one-dimensional hierarchy based on the mean education and earnings of each occupation’s incumbents, and it is claimed to provide a good proxy for long-term economic status, even better than single-year income or earnings [Goldberger Reference Goldberger1989; Zimmerman Reference Zimmerman1992]. Several updates and refinements of the original occupational status index recalibrate the weights attached to education and income, account for the upgrading of occupational structure and for gender differences in occupational participation [Stevens and Featherman Reference Stevens and Featherman1981; Nakao and Treas Reference Nakao and Treas1994], and offer an internationally comparable index [Ganzeboom et al. Reference Ganzeboom, De Graaf and Treiman1992]. These measures have produced an important body of work (e.g. Grusky and DiPrete Reference Grusky and DiPrete1990, Beller and Hout Reference Beller and Hout2006, Yaish and Anderson 2012). However, the sociological analysis of occupational status mobility has declined as earnings and income mobility have gained centrality.
While measures of occupational status subsume all sources of socioeconomic advantage into a single one-dimensional hierarchy, classes are categorical groupings based on specific occupational assets believed to determine life chances [Wright Reference Wright1985; Reference Wright1997; Goldthorpe Reference Goldthorpe1987; Grusky and Weeden Reference Grusky, Weeden, Morgan, Grusky and Fields2006]. Classes capture not only the amount but also the type of occupational resources controlled by individuals, and thus considers “the sources of inequality and not only its surface manifestations” [Portes and Hoffman 2003]. Researchers claim that different classes are distinctly affected by contextual factors such as economic restructuring, globalization, or labor market policies, even if their incumbents have similar levels of occupational status [Breen and Whelan Reference Breen and Whelan1996]. As such, differences between classes are not reduced to a single hierarchy.
Class is not just an economic phenomenon. Similar life chances may provide a structural platform for the formation of common values, attitudes, tastes, and behaviors, which will define “classes as cultures” [Wright Reference Wright1997; Willis Reference Willis1981; Bourdieu Reference Bourdieu1984; Lamont Reference Lamont1992; Reference Lamont2002]. Furthermore, members of a class can gain awareness of their common interests and build a shared identity, which can become the basis of collective action. This defines “classes as actors,” as expressed in the famous Marxist “class for itself” [Marx Reference Marx and Fernbach1973: 238-239]. Even though the notion of classes as cultures and actors is important for understanding the impact of stratification on social dynamics, mobility analysts focus on the Weberian “weak” notion of classes [Wright Reference Wright2002], as determinants of “unequal chances individuals have of sharing in the economic and cultural goods which exist in society” [Giddens Reference Giddens1973: 130-131] i.e. on classes as expression of socioeconomic inequality.
The most widely used class classification was devised by Erikson, Goldthorpe and Portocarero [Reference Erikson, Goldthorpe and Portocarero1979]. The classification is based on “employment relations” [Erikson and Goldthorpe Reference Erikson and Goldthorpe1992: 35-47; Goldthorpe Reference Goldthorpe2007: 101-124] and combines the distinction between employer, self-employed and employees with further distinctions based on level of skill, sector of employment, and authority in the workplace. Its most common version has seven classes, including professionals and managers, clerical workers, self-employed, farmers, skilled manual workers, and farm workers. This class classification emerged in the context of a major project known as the Comparative Analysis of Social Mobility in Industrialized Countries casmin, which compared mobility across fifteen industrialized countries in the 1970s. An alternative class classification was devised by Wright [Reference Wright1985; Reference Wright1997]. This perspective expands the Marxist notion of exploitation by defining “multiple exploitations” based on ownership of the means of production (capital), organizational assets (authority) and skills (expert knowledge) as sources of class divisions. Both class schemas have been used in empirical research but the casmin classification has become the standard for comparative mobility analysis.
Earnings and income mobility
Economists favor pecuniary measures of economic well-being, including earnings and income. Like class and status, earnings provide a measure of well-being strictly based on the labor market. As a result, they do not include those who are not working or extra-occupational resources, such as financial assets and public and private transfers. They thereby potentially exclude the “underclass,” poorly attached to the labor market [Grusky and Weeden 2008], and the “overclass,” whose income largely depends on returns to capital. Total family income includes returns from labor earnings, assets, and transfers accruing to all family members. By focusing on the family, income measures account for dynamics such as spousal selection (assortative mating) and intra-household division of labor that are neglected in individual-level measures. Income data have several advantages: in addition to capturing diverse sources of revenue, income measures have a one-to-one correlation with standard measures of economic inequality. Furthermore, high-quality intergenerational income data are increasingly available from tax registries and other administrative sources. These factors explain the growing relevance of income in the study of mobility [Chetty et al. Reference Chetty, Hendren, Kline and Saez2014; Mitnik et al. Reference Mitnik, Bryant, Weber and Grusky2015].
How is mobility measured and what are the main empirical findings of mobility research?
The strategies to capture intergenerational mobility depend on the measure of status used. With continuous variables such as occupational status, income, and earnings, researchers rely on linear regression models that capture the expected value of children’s status across levels of parental status. Because income and earnings distributions are highly skewed, the variables are recoded using logarithms, which transform the regression coefficient into an elasticity indicating the proportion of a one percent difference in parents’ income that will be transmitted as income differences between their children. Measures of elasticity usually range between zero and one, with zero indicating no intergenerational association and one indicating that the proportional earnings differences of parents will be exactly replicated in the children’s generation [Blanden Reference Blanden2013].
Empirical estimates of earnings and income elasticities are currently available for about two dozen countries. They vary widely between approximately 0.15 and 0.70. In the United States, the country on which probably the most research exists, estimates of income and earnings elasticities range between 0.30 and 0.60 depending on methodological decisions [Black and Devereux Reference Black, Devereux, Card and Ashenfelter2011].
Elasticities, the staple measure of intergenerational persistence, include information about both the association between parents and children and any changes in the income distribution across generations. Ceteris paribus, an increase in inequality across generations will increase the elasticity, and a decline in inequality will reduce it. To control for changes in inequality, researchers have used alternative measures such as the correlation coefficient and the rank-rank coefficient linking parents’ and children’s income ranks rather than levels. The choice of measure is consequential in contexts that have experienced substantial changes in inequality. For example, intergenerational persistence is stronger in the US than in Sweden and the UK when using elasticity but similar using the rank-rank correlation or the correlation coefficient [Corak et al. 2014; Eberharter 2013], suggesting that the “mobility deficit” in the US has been exacerbated by growing inequality.
Naturally, reducing the mobility process to a single measure of linear average persistence across generations is a simplification. Recent work uses alternative strategies to capture the pattern and not just the overall level of intergenerational association. These include matrices cross-classifying parents’ and children’s income quintiles, models allowing for nonlinearities in the intergenerational association, and quantile regression models examining income dispersion around the regression line [Corak and Heisz Reference Corak and Heisz1999; Couch and Lillard Reference Couch, Lillard and Corak2004; Peters Reference Peters1992; Eide and Showalter Reference Eide and Showalter1999]. Based on these approaches it is possible to examine whether average persistence is stronger for the rich or the poor, and whether the dispersion in children’s income varies across the socioeconomic hierarchy. For example, Corak and Heisz (Reference Corak and Heisz1999) find that intergenerational persistence is greater at the upper end than at the lower end of earnings distribution in Canada, and Eide and Showalter [Reference Eide and Showalter1999] found much more variance around predicted children’s income for poor than for wealthy parents in the US. Both findings signal stronger persistence at the top.
An important challenge in the analysis of economic mobility is that single-year measures of earnings or income provide poor proxies for long-term economic status. The problems are manifold. First, respondents are reluctant to report their income, likely to report it with error, and likely to experience transitory income fluctuations that can persist over many years (measurement error and persistent transitory shocks). Furthermore, the transitory component of income varies over the life cycle (age-related bias), and the association between current and long-term earnings is not constant over the life cycle (life-cycle earnings bias). Starting in the 1980s, a small cottage industry in economics has addressed these issues [Jenkins Reference Jenkins1987; Zimmerman Reference Zimmerman1992; Solon 1992; Peters Reference Peters1992; Mazumder Reference Mazumder2005, Haider and Solon Reference Haider and Solon2006; Grawe Reference Grawe2006]. Currently, the convention is to take income averages over many years in order to reduce measurement error, and to center measures around age 40, a stage in the life cycle that has been found less prone to bias.
A recent alternative attempt to address these measurement challenges uses surname-level averages of economic status for parents and children [Clark Reference Clark2014; Clark et al. Reference Clark, Cummins, Hao and Vidal2015]. Such averages are claimed to capture individuals’ underlying competence purged from random variation. Intergenerational persistence of surname-level status is very high—with coefficients of around 0.80 or even higher—and surprisingly stable across time and place, including medieval England, Sweden since 1600, and Chile in the 20th century. Based on these findings, Clark [Reference Clark2014] argues that intergenerational mobility is limited, impervious to social policy and that, because its main determinant is exogenous genetic endowments shaping competence, it should be left alone. The approach suffers from two serious problems. First, surname-level averages capture not only individual-level attributes but also group-level factors—e.g. discrimination, group networks, historical barriers to integration, etc.—that have an independent effect on individual status (hence the high estimates of persistence). Second, taking group-level averages does not solve the measurement error issue that motivates their use [Torche and Corvalan Reference Torche and Corvalan2015]. Surname data can provide interesting information for historical analyses when individual-level data are not available, but it does not provide an adequate measure of mobility as experienced by individuals.
In sum, the analysis of economic mobility involves substantial methodological challenges. These methodological issues are not just a technical detail. Rather, they have vast implications for the estimated magnitude of the intergenerational association. In the US for example, early measures of intergenerational earnings elasticities yielded low values of 0.15-0.20 [Becker and Tomes Reference Becker and Tomes1986; Behrman and Taubman Reference Behrman and Taubman1985], leading to the conclusion that earnings were not strongly transmitted across generations. As concluded by Becker and Tomes [Reference Becker and Tomes1986: 32], “aside from families victimized by discrimination […] almost all earnings advantages and disadvantages of ancestors are wiped out in three generations.” We now know that this finding was an artifact of using single-year measures of earnings and small, homogenous samples. The use of better techniques has led to the consensual figure of intergenerational elasticity being raised to about 0.40 in the 1990s [Solon Reference Solon, Ashenfalter and Card1999], and to around 0.50 or more in the 2000s [Black and Devereux Reference Black, Devereux, Card and Ashenfelter2011; Torche Reference Torche2015]. Seemingly small methodological decisions can have large consequences. For example, decisions about how to deal with non-linearities and with zero income values in the US result in estimates of the intergenerational elasticity widely varying between 0.30 and 0.70, which spans most of the estimates in the empirical literature [Chetty et al. Reference Chetty, Hendren, Kline and Saez2014].
After decades of research, we have learned that producing reliable estimates of economic mobility is a colossal task. The growing availability of administrative data with little measurement error and comprehensive coverage of the life-course should alleviate the problem. Still, no single estimate mobility will provide a conclusive answer. Rather, it should be seen as an additional data point in the distribution of plausible measures.
The analysis of class mobility uses altogether different techniques. Given that differences between classes cannot be subsumed to a single hierarchy, analysts consider the entire mobility table cross-classifying classes of origin and destination, and model the associations in different regions of the table with log-linear models [Featherman and Hauser 1978; Hout 1984]. The landmark casmin project comparing advanced industrial countries found that countries differ in their occupational structure, but net intergenerational mobility was extremely similar across countries and over time. These findings were expressed in the “common” and “constant” social fluidity hypotheses, respectively. The basic temporal and international similarity was systematized into a “core model” [Erikson and Goldthorpe Reference Erikson and Goldthorpe1992]. The model is theoretically driven and distinguishes mobility barriers based on hierarchy (status differences between classes), inheritance (class-specific propensities to remain in classes of origin) and sector (barrier between agricultural and non-agricultural classes), in addition to country-specific affinities or disaffinities between specific classes. This model has been the obligatory referent for comparative class mobility research, but it has been subject to important criticism [Hout and Hauser Reference Hout and Hauser1992].
Analysis of country-specific trends in class mobility has questioned the “constant fluidity hypothesis.” A major follow-up of the casmin project analyzing mobility until the 1990s in 11 European countries found growing mobility in some countries, but null or slight temporal change in others [Breen 2004]. While Britain, Israel and Germany are best characterized by “constant fluidity”; growing openness is detected in France, Hungary, Ireland, Italy, the Netherlands, Norway, Poland, and Sweden. Changes in Ireland and Italy are quite minor [Layte and Whelan Reference Layte, Whelan and Breen2004; Pisati and Schizzerotto Reference Pisati, Schizzerotto and Breen2004], and France, the Netherlands, and Brazil display a sustained increase in mobility [Ganzeboom and Luijkx 2004; Vallet Reference Vallet and Breen2004; Torche and Ribeiro Reference Torche and Ribeiro2010]. An important open question is the contextual factors accounting for variation in class mobility across time and place.
Mobility and inequality
If mobility captures equality of opportunity, a central concern is how mobility relates to cross-sectional inequality. Much literature suggests that higher inequality reduces intergenerational mobility. Several theoretical mechanisms account for this relationship. Inequality in the parental generation implies wider disparities in the formative environments of children and less progressive human capital investments, which results in less mobility [Neckerman and Torche Reference Neckerman and Torche2007; Ermisch et al. Reference Ermisch, Jantti, Smeeding, Wilson, Ermisch, Jäntti and Smeeding2012]. Higher inequality also implies higher returns to schooling in labor and other markets, which induce intergenerational rigidity [Solon Reference Solon and Corak2004]. Additionally, inequality may induce residential segregation, resulting in a more skewed composition of peer groups along socioeconomic lines [Durlauf 1996; Reardon and Bischoff Reference Reardon and Bischoff2011].
An alternative pathway of influence is the political realm. Political dynamics can result in a positive or a negative link between inequality and mobility. As suggested by the median-voter theorem, higher inequality induces median voters to push for redistribution, which could foster mobility. Evidence about this pathway is inconsistent. Some cross-national studies find support for the median voter hypotheses [Kenworthy and Pontusson Reference Kenworthy and Pontusson2005; Milanovic Reference Milanovic2000]. In others the association is null or negative [Perotti Reference Perotti1996] or varies depending on the type of redistributive program [Osberg et al. Reference Osberg, Smeeding, Schwabish and Neckerman2004]. The political system could also result in a negative association between inequality and mobility if high economic concentration strengthens the influence of the wealthy through political contributions and lobbying, thus reducing the scope for redistributive policies [Burtless and Jencks Reference Burtless, Jencks, Henry, Lindsay and Nivola2003]. No conclusive evidence about this hypothesis exists to date. Research in the United States has shown that inequality matters for political influence [Bartels Reference Bartels2008; Gilens Reference Gilens2012], but it is not clear whether the pattern of unequal influence has increased as inequality has risen, and no comparable evidence exists in other countries [Kenworthy Reference Kenworthy2015].
These mechanisms linking inequality and mobility are plausible, but they leave much to be explained in terms of their level of analysis, appropriate time lags, first-order effects versus externalities, potential tipping points, and the vast risk of spuriousness. As we will discuss, the association between inequality and mobility may largely be driven by other variables that shape both dimensions of inequality, such as institutional arrangements and welfare state configurations.
Inequality and mobility: the evidence
Surprisingly, cross-country comparisons of class mobility show a small or null association between mobility and economic inequality [Erikson and Goldthorpe Reference Erikson and Goldthorpe1992; Breen and Luijkx Reference Breen, Luijkx and Breen2004; Torche 2005; Ribeiro Reference Ribeiro2007]. This finding again suggests that social class captures sources of advantage distinct from economic disparities. In contrast, there is a strong positive association between cross-sectional inequality and intergenerational earnings and income persistence. Egalitarian Scandinavian countries feature the highest levels of mobility with elasticities of about 0.20, whereas more unequal Italy, the United Kingdom, and particularly the United States are less mobile with elasticities reaching 0.40 or more [Bjorklund and Jantti Reference Bjorklund, Jantti, Salverda, Nolan and Smeeding2009; Blanden Reference Blanden2013; Jantti et al. Reference Jantti, Bratsberg, Roed, Raaum, Naylor, Osterbacka, Bjorklund and Eriksson2006; Solon Reference Solon2002; Jantti and Jenkins Reference Jantti, Jenkins, Atkinson and Bourgignon2015]. When much more unequal Latin American countries have been added to the comparison, they feature even higher elasticities reaching about 0.60 or more [Torche Reference Torche2014]. The negative cross-country association between inequality and mobility has transcended academia and been popularized in the “Great Gatsby” curve [Corak Reference Corak2013; Krueger Reference Krueger2012].
The association between economic inequality and earnings mobility fails, however, to materialize when changes within countries are examined. The most salient case is the United States, which has experienced large increases in economic and educational disparities since the late 1970s, with the household income Gini rising from 0.40 in 1980 to 0.48 in 2012 [US Census Bureau 2015], and access to higher education becoming more stratified by income [Belley and Lochner Reference Belley and Lochner2007]. However, mobility does not seem to have declined during this period, with the large majority of studies finding insignificant change over time [Fertig Reference Fertig2003; Mayer and Lopoo Reference Mayer and Lopoo2005; Hertz Reference Hertz2007; Lee and Solon Reference Lee and Solon2009; Bloome Reference Bloome2014]. While this inconclusiveness could be attributed to data limitations, a recent analysis using a large administrative dataset with little measurement error suggests that the absence of mobility trends as inequality has increased may be an accurate finding [Chetty et al. Reference Chetty, Hendren, Kline and Saez2014].
Trends in other countries are more consistent with a negative association between inequality and mobility, but the evidence is weak. The United Kingdom experienced a large increase in inequality between 1980 and 1990 [Equality Trust 2011]. Researchers have found an increase in the intergenerational elasticity for cohorts born between 1960 and 1970, but not between those born between 1950 and 1960 [Blanden et al. 2013; Nicoletti and Ermisch 2007], a trend that does not fully match with the increase in inequality. Another interesting case is Finland, in which an increase in economic mobility was detected from cohorts born between 1930 and 1950, a period in which inequality largely declined [Pekkala and Lucas Reference Pekkala and Lucas2007]. However, growing mobility may have been entirely driven by comprehensive school reform, without the additional influence of income inequality [Pekkarinen et al. Reference Pekkarinen, Uusitalo and Kerr2009].
The strong association between inequality and mobility across countries but weak association in their trends within countries strongly suggests that the link between inequality and mobility is spurious, driven by institutional arrangements organized into welfare state regimes [Esping-Andersen Reference Esping-Andersen1990; Hall and Solstike Reference Hall and Soskice2001; Arts and Gelissen Reference Arts and Gelissen2002]. Coherent configurations of institutional arrangements, including redistribution via taxes and transfers, protection against risk, and early investments in human capital likely shape both cross-sectional inequality and its transmission across generations in ways that vary across countries, but which result in limited immediate change within countries when inequality drops or rises in the short-term.
Studying the influence of such institutional arrangements on intergenerational persistence is a challenging task. The most common strategy involves comparing countries exploiting their variation in institutional arrangements [Ichino et al. Reference Ichino, Karabarbounis and Moretti2011; Behrman et al. Reference Behrman, Birdsall, Szekely, Birdsall and Graham1999; Beller and Hout Reference Beller and Hout2006]. For example, Ichino et al. [Reference Ichino, Karabarbounis and Moretti2011] found a strong correlation between public expenditures on education and intergenerational income elasticity across ten advanced industrial countries. This cross-country multivariate strategy is limited, however, by small sample sizes and high correlation between institutional and economic factors, preventing claims beyond the thinnest of description.
An alternative strategy is to exploit plausibly exogenous variation in specific policies to assess their causal effect on mobility. For example Pekkarienen et al. [Reference Pekkarinen, Uusitalo and Kerr2009] exploited the fact that comprehensive school reform was implemented gradually across the country in Finland to examine its effect on intergenerational income elasticity. Similarly, Rauscher [forthcoming] leveraged variation across time and state in early compulsory schooling laws in the US to assess the causal effect of educational expansion on mobility. The main limitation of this isolating policy effects strategy is that examining a single policy component may provide a correct but hardly generalizable answer if the effect depends on the institutional milieu in which it emerges. While neither strategy alone provides a conclusive answer, their integration offers a promising avenue to explain the determinants of mobility.
Mechanisms of mobility
A long tradition in sociology has examined the mechanisms that account for the transmission of advantage across generations. The seminal work by Blau and Duncan [Reference Blau and Duncan1967] pioneered the status attainment tradition by offering a mobility model including its intergenerational and life-cycle components. The key variables in this model are father’s education and occupational status, children’s educational attainment, children’s occupational status at first job, and adult children’s socioeconomic status. The examination of mechanisms was continued by the Wisconsin school, which added psychosocial mediating factors including cognitive ability, significant others’ influences, aspirations, expectations, and school performance to the status attainment model [Sewell et al. Reference Sewell, Haller and Portes1969; Reference Sewell, Haller and Ohlendorf1970; Haller and Portes Reference Haller and Portes1973]. Mobility mechanisms have recently been rediscovered by economists, who use similar multivariate models as those in the status attainment tradition and include mediators such as educational attainment, cognitive skill, personality traits, and occupational experience (e.g., Eide and Showalter Reference Eide and Showalter1999; Blanden, Gregg, and Macmillan Reference Blanden, Gregg and Macmillan2007, Bowles, Gintis, and Groves Reference Bowles, Gintis and Groves2005; Smeeding, Erikson, and Jantti Reference Smeeding, Erikson and Jantti2011).
One important finding in the tradition accounting for mechanisms is the central role of education as both the main vehicle for intergenerational reproduction and the main avenue for mobility [Hout and DiPrete Reference Hout and DiPrete2006]. Education is the main vehicle for reproduction because most of the intergenerational association is mediated by children’s educational attainment. Three components of its mediating role are conventionally analyzed: the association between social origins and educational attainment (conventionally called “inequality of educational opportunity”), the association between education and socioeconomic advantage (“returns to schooling”) and the direct intergenerational association once education is accounted for.
The direct intergenerational association net of education has been found to be non-zero but minor [Blau and Duncan Reference Blau and Duncan1967; Sewell and Hauser Reference Sewell and Hauser1975], and to vary across levels of educational attainment. It is strong among those with less than a college degree, very weak among college graduates, and strong again among those with a graduate degree [Hout Reference Hout1988; Torche Reference Torche2011]. While a college degree fulfills its meritocratic promise, the strong intergenerational association among those with a graduate degree questions the meritocratic nature of highly skilled labor markets, and refocuses attention on features of the higher-education system that contribute to intergenerational persistence by providing avenues for advantaged families to invest in their children’s schooling. As post-secondary educational systems have expanded and diversified, important sources of such “horizontal educational stratification” may include institutional selectivity and field of study (Gerber and Cheung Reference Gerber and Cheung2008). The weak intergenerational association among college-degree holders also offers an avenue for increasing mobility through educational expansion via a compositional effect: as education expands and larger proportions of each cohort gain access to a college degree, mobility should increase due to the growing numbers of college graduates in the population. This compositional effect has contributed to mobility in several countries including Sweden, Great Britain, Germany, France, and the United States [Breen and Jonsson Reference Breen and Jonsson2007; Breen Reference Breen2010; Vallet Reference Vallet and Breen2004; Pfeffer and Hertel Reference Pfeffer and Hertel2015]. An open question is whether the meritocratic power of the college degree is partly driven by positive selectivity of college graduates and may decline as college expansion reduces selectivity.
An important sociological contribution is the comparative analysis of inequality of educational opportunity across time and place. Early work comparing thirteen advanced industrial countries found persistent educational inequality in spite of massive educational expansion [Shavit and Blossfeld Reference Shavit and Blossfeld1993] but more recent analysis has found equalization in several countries (Breen et al. Reference Breen, Luijkx, Müller and Pollak2009). Beyond self-evident financial constraints, inequality of educational opportunity has been accounted for by class-stratified choices. Boudon [Reference Boudon1974] argues that class differences in educational attainment depend on prior academic performance and choice conditional on performance, in particular the choice to continue in school at each educational transition; and that the influence of past performance on choice is class-stratified. Disadvantaged children are highly sensitive to prior performance when deciding whether to continue to the next educational level while advantaged children persist regardless of their prior attainment. Cross-country variation in inequality of educational opportunity appears to emerge largely from class-based choices, not academic performance [Jackson and Jonsson Reference Jackson, Jonsson and Jackson2013].
The search for mobility mechanisms is a laudable enterprise but it faces theoretical and empirical limitations. At the theoretical level, both status attainment and economic traditions are largely oblivious of structural factors, and in particular power dynamics, in explaining the role of the educational system in social mobility. Critical approaches argue that the educational system may serve as an institutional device for intergenerational class persistence (Bourdieu Reference Bourdieu, Karabel and Halsey1977a; Reference Bourdieu1977b, Bourdieu and Passeron Reference Bourdieu and Passeron1977; Bowles and Gintis Reference Bowles and Gintis1976). As elaborated by Bourdieu, schools may provide a vehicle to legitimize and maintain the socioeconomic structure by transforming class distinctions into educational distinctions represented as emerging from merit, and channeling children of different social origins into different positions. These critical approaches remind us of the limits of educational expansion as a strategy to foster mobility.
At the empirical level, the search for mechanisms in both sociology and economics relies on path-analytic models. Researchers add mediating variables to a model linking parents’ and children’s status and interpret the path connecting origins to destinations through each putative mediator as capturing their mediating role. The coefficient associated with parental status once mediators are added is interpreted as the “direct effect” of social origins. This is not a correct interpretation. Virtually all putative mediators are endogenous and they provide limited information about mechanisms for transmission. To provide an intuitive interpretation of this claim, consider the following: accounting for a child’s educational attainment as a mediator means comparing children with the same level of educational attainment but different social origins: say, two college graduates, one from a poor family and one from a wealthy family. If educational attainment is correlated with unobserved factors shaping economic success (for example, cognitive ability or motivation), the model implies comparing a poor child who “made it to college” (perhaps given her unobserved outstanding ability) with a wealthy child who “just attained a college degree” (perhaps given her limited motivation). This inappropriate comparison of their potential outcomes departs from a causal effect. If we could account for all the factors that account for children’s socioeconomic status and that may be passed across generations, we would be able to rule out endogeneity and parameter estimates would provide information about causal connections. But such exhaustive measurement of all potential pathways of intergenerational influence is unrealistic.
Alternatively, researchers could exploit random variation in putative mediators, such as by using natural experiments, in order to capture their causal effect. But the analytical requirements for causal interpretation of mechanisms are stringent. Technically, mediation analysis requires “sequential ignorability” i.e. both parents’ status and the putative mediator should be allocated at “as random” [Imai et al. Reference Imai, Keele, Tingley and Yamamoto2011]. This is certainly not the case in naturalistic settingsFootnote 2.
Scholars are increasingly aware of these challenges to causal interpretation. A common response is to claim that even if no causal effects can be captured, mobility models including mediators are valuable descriptive devices that help researchers assess the different pathways for the transmission of advantage. The problem with this approach is that descriptive models can yield estimates that differ not only in magnitude but also in size from causal estimates. This is a concern in terms of theoretical or policy implications of the findings. Description has a central place in the study of mobility, but given the challenges of establishing causal relationships from observational data when multiple variables are included and mediation is assumed, it is probably prudent to focus the descriptive effort on the bivariate intergenerational association.
Multigenerational mobility
Another application of the question about mechanisms for mobility is the study of intergenerational associations across more than two generations. Although the analysis could consider many generations of ancestors, most of this research has expanded to include three generations––grandparents, parents, and children [Mare Reference Mare2011]. In these settings, the parental generation can be seen as a mediating factor linking grandparents to children. The initial question when mobility across multiple generations is considered is whether the process is Markovian, i.e., one in which the outcome at any stage depends only on the outcome of the previous stage. In this “memoryless” process, the status of children depends on their parents’ status only, without an additional direct role for grandparents or other ancestors.
Empirical analysis of three-generation association dates back to the 1950s and 1960s [Svalastoga Reference Svalastoga1959; Hodge Reference Hodge1966]. In a classic study, Hodge [Reference Hodge1966] found that the process of mobility was largely Markovian, with little direct association between grandparents and children. The finding of null or weak grandparents’ effects once parental status is accounted for has been replicated with different measures of economic well-being [Hauser and Warren Reference Hauser and Warren1997; Erola and Moisio Reference Erola and Moisio2007], although some studies have found a direct association between the status of grandparents and children [Chan and Boliver Reference Chan and Boliver2013; Lindahl et al. Reference Lindahl, Palme, Massih and Sjogren2012].
An important challenge for the analysis of multigenerational mobility is the interpretation of the direct association between grandparents and children net of parents’ resources. This problem is just another example of the limitations of mediation analysis with observational data. Theoretical models in economics have indeed predicted a small negative association between grandparents and grandchildren when the association is conditional on children’s status [Becker and Tomes Reference Becker and Tomes1979, Solon Reference Solon2014]. As put by Solon [Reference Solon2014] “if the parents did not earn more despite the advantages of higher grandparents’ income, this signals that the parent got a poor draw in the genetic/cultural endowment, and that poor draw tends to be passed to some extent to children.” The intuition behind this example suggests that when we condition on parents’ socioeconomic status, we compare children whose parents have similar status—say, low—but whose grandparents have different statuses and who are therefore very likely to differ in terms of unobserved factors (what Solon calls genetic/cultural endowments). Controlling for parents’ status can result in a “grandparents’ effect” that differs not only in magnitude but also in sign from the true causal effect.
Given these challenges, a productive avenue might be to examine specific mechanisms linking more than two generations using a falsification logic. For example Zeng and Xie [Reference Zeng and Xie2014] reasoned that if grandparents shape children’s outcomes through interacting with them, then their influence should be stronger when they co-reside with their grandchildren than when they live somewhere else or are deceased. Their analysis of rural China confirms that indeed the education of co-resident grandparents has a positive association with grandchildren’s education, but the education of deceased or non-co-resident grandparents does not. This dual finding strongly supports the plausibility of an interactional mechanism. More examination of theoretically-driven mechanisms is needed to advance the understanding of influences of grandparents and extended families on individual well-being.
Does intergenerational mobility measure equality of opportunity?
We care about mobility in large part because it provides information about equality of opportunity. But perfectly equal opportunity does not imply eliminating all sources of socioeconomic resemblance between parents and children. Equal opportunity does not require, nor can it request, equalizing factors such as inherited differences in ability and early household socialization of tastes [Jencks and Tach Reference Jencks, Tach, Morgan, Grusky and Fields2006]. Assuming that these factors play a role in every society, equal opportunity will not result in a null association between parents and children.
The extent to which mobility does capture equality of opportunity is claimed to depend on which mechanisms account for intergenerational reproduction [Jenck and Tach Reference Jencks, Tach, Morgan, Grusky and Fields2006; Swift Reference Swift2004]. To the extent that persistence emerges from class-based achievement barriers preventing children from developing their full potential due to lack of parental access to financial, cultural or social resources, or preventing adults from being rewarded for their productivity due to factors transmitted across generations such as racial discrimination, the intergenerational association will be interpreted as capturing inequality of opportunity. But if intergenerational persistence is due to genetic inheritance or by cultural endowments transmitted through early socialization of tastes, it is interpreted as unrelated to equality of opportunity.
This argument is problematic. As section 4 argues, it is difficult to disentangle the causal effect of different mechanisms. More importantly, a growing body of research from the biological and behavioral sciences shows that the distinction between barriers to achievement and inheritance of endowments such as genes or tastes exists only at the analytical level and is probably immaterial for practical purposes. Research shows that individual capabilities are shaped by the socioeconomic environment starting as early as the prenatal period, in multiple ways that have consequences for later life, shaping health, development, and socioeconomic well-being [Knudsen et al. Reference Knudsen, Heckman, Cameron and Shonkoff2006; Palloni Reference Palloni2006]. As a landmark study summarizes, “virtually every aspect of early human development is affected by the environment and experiences encountered in cumulative fashion, beginning in the prenatal period and extending into the early childhood years” [Shonkoff and Phillips Reference Shonkoff and Phillips2000: 6].
The same may apply to so-called cultural endowments transmitted through early socialization. As elaborated by Bourdieu [Reference Bourdieu1977b: 82-83] social class constraints shape tastes and dispositions from early on in life in ways that are unconscious, coalescing into a “system of lasting, transposable dispositions which, integrating past experiences, functions at every moment as a matrix of perceptions, appreciations, and actions.” This approach suggests that attributes usually described as temperament or propensities may be firmly rooted in class circumstances, and questions the exogeneity of tastes.
Social class also shapes endowments in less subtle, more blatant ways. Disadvantaged children are more likely to be exposed to a myriad of environmental insults such as violence [Harrell et al. 2014], lead [Tong et al. 2000], and pollution [Bell and Ebisu 2012], which may severely constrain their cognitive development in early life. Recent research using neuroimaging technology suggests that poverty is associated with brain structure and function very early in childhood, and this association is largely net of genetic variation [Noble et al. Reference Noble2015; Hair et al. Reference Hair, Hanson, Wolfe and Pollak2015]. A recent scholarship on epigenetics further transcends the distinction between socioeconomic constraints and transmission of endowments. Epigenetics (meaning “above” or “in addition to” genes) shows that environmental factors alter gene expression and activity in ways that are persistent and heritable but which do not involve change in dna sequence. For example, prenatal exposure to stress––which is highly stratified by socioeconomic status [Turner et al. Reference Turner, Wheaton and Lloyd1995]—results in low birth weight driven by gene expression of “stress hormones” in the placenta [Hobel Reference Hobel2004; McLean et al. Reference McLean, Bisits, Davies, Woods, Lowry and Smith1995]. Given that birth weight predicts health, development and socioeconomic outcomes [Conley et al. Reference Conley, Strully and Bennett2003], this epigenetic effect shapes individual endowments in ways that could easily be interpreted as hereditary but that are purely environmental. Because epigenetic influences can be inherited even if there is no change in gene structure [Yehuda et al. 2015], the effects of the environmental exposures can be transmitted across generations, much as dna can.
These bodies of research naturally do not rule out the potential for an autonomous role of genes or tastes in the transmission of advantage. However, they suggest that environmental exposures shape individual health and attainment from conception in durable ways that can easily, and incorrectly, be construed as genetic or cultural endowments. This renders the distinction between endowments and socioeconomic constraints blurry to the point of being immaterial. It also suggests that policies intended to promote upward mobility among disadvantaged populations are necessary but not sufficient to alter opportunity in a single generation, and gives credence to the longstanding insight that a potent way to equalize intergenerational opportunity is to equalize economic well-being across families, particularly those with children (e.g. Jencks et al. Reference Jencks, Smith, Acland, Jo Bane, Cohen, Gintis, Heyns and Michelson1972).
Summary and conclusions
Intergenerational mobility, operationalized as the strength of the association between parents’ and adult children’s socioeconomic status, is a central concept for stratification research because it provides information about inequality of opportunity in society. Producing a valid and reliable measure of intergenerational association is no small feat. It requires selecting a measure of status that captures long-term economic well-being purged from temporary fluctuation at the right stage of the life course for both parents and children. These challenges are not just a methodological detail. Rather, they have major substantive consequences affecting the assessment of mobility across countries and over time.
Different measures of intergenerational association exist, including class, occupational status, education, income, earnings, and wealth. Sociological scholarship has historically focused on class and occupational mobility and has produced a valuable body of research. However, recent academic and policy discussion emphasizes economic mobility measured by earnings and income. Several reasons explain the emphasis on income: the simplicity in describing the intergenerational association through a single parameter linking continuous variables, the one-to-one correlation between metrics for mobility and economic inequality, and the growing availability of income and earnings data from administrative sources.
Research on economic mobility shows substantial variation across countries and a strong association between intergenerational economic persistence and cross-sectional economic inequality. In one extreme, egalitarian Scandinavian countries feature low levels of intergenerational persistence. In the other, highly unequal Latin American countries feature very strong persistence. The United States features both the highest level of inequality and the lowest level of mobility among affluent nations. In contrast, mobility trends within countries are only weakly—if at all—related to changes in inequality. This inconsistency strongly suggests that the mobility-inequality relationship is spurious, likely driven by institutional arrangements shaping the distribution of opportunity and rewards through policy regimes including taxes and transfers, protection against risk, and early investments in human capital. Spuriousness does not mean that cross-sectional inequality is unimportant. Rather, it moves the analytical and policy question one step prior, to the institutional configurations shaping life chances. It also suggests that short-term changes in economic inequality or narrow interventions will have limited influence on mobility.
Researchers have extended the study of mobility beyond its bivariate focus by examining mediators of the intergenerational association. Particular emphasis has been placed on educational attainment, cognitive ability, personality traits, aspirations, and school performance. These attempts are problematic because putative mediators are correlated with unobserved determinants of economic well-being, making the models uninformative in terms of causal processes. While description is an important task in the social sciences, and fixation with causal identification might favor identifiable but irrelevant questions, the parameter estimates of models including mediators may depart from true causal effects not only in terms of magnitude but also in terms of sign. This is a major concern for the analytical and policy implications of the analysis and suggests that it is probably prudent to focus the descriptive effort on the bivariate intergenerational association.
The main reason we care about mobility is because it captures equality of opportunity. However, it has been argued that not all sources of socioeconomic resemblance between parents and children signal unequal opportunity. Intergenerational persistence emerging from genetic transmission or tastes/preferences socialization will continue to exist even in a perfectly fair society, and thus perfect mobility does not require a zero intergenerational association. This view, even if correct at the analytical level, is questioned by growing evidence suggesting that endowments emerging early in life are critically shaped by the socioeconomic environment. In particular, recent developments in neurobiology and epigenetics show that attributes considered inherited are indeed shaped by socioeconomic circumstances from conception. These discoveries do not rule out the potentially autonomous role of genetic or cultural transmission. However, they do suggest that truly exogenous transmission of endowments might be both slight and similar across time and place, and that the comparative analysis of mobility does inform us about inequality of opportunity in different societies.
The analysis of mobility is still a relatively young enterprise, and it is still largely restricted to the affluent world. Much has been advanced over the last few decades in terms of methodological strategies to produce a valid and reliable measure of intergenerational persistence. The standardization of methods and the growing availability of mobility data will likely further the understanding of mobility around the world in the future. However, it is fair to say that mobility analysis is empirically strong but theoretically weaker. As empirical progress is made, more theoretical work delineating the contours, limits, and implications of mobility analysis is needed.
Acknowledgements
The author would like to thank Dalton Conley, Vida Maralani and members of the editorial board of European Journal of Sociology for their helpful comments and suggestions.
