American geography of opportunity reveals European origins

Our selection of European populations is guided by available cross-country data on income mobility. Corak (6) reports such estimates for 13 countries; 8 of these are European, and 7 have an ancestral population in the United States of at least 1 million. Ordered by decreasing size of the US population, these are: the United Kingdom, Germany, Italy, France, Norway, Sweden, and Denmark. Given the cultural overlap between the Scandinavian countries, we collapse them throughout much of our analysis. We use individual-level data on self-identified ancestry from the 1980 US Census to estimate the share of the population with these European origins for each of the 722 commuting zones (CZs), which are local labor markets that exhaust the contiguous United States (Materials and Methods).

We focus on three different CZ-level measures of income mobility (Materials and Methods). (i) The first is upward mobility—the expected income percentile rank in adulthood for a child born to parents in the bottom half of the distribution (4). We use separate estimates for blacks and non-Hispanic (non-Hisp.) whites and also examine a complementary measure: “rags-to-riches” mobility, which captures the probability that a child born in the bottom income quintile ends up in the top (4, 8). (ii) For comparability with cross-country data, we also examine the intergenerational rank correlation, which measures persistence throughout the distribution rather than mobility out of the bottom (4). (iii) We use estimates of the causal place effect of each CZ on upward mobility, addressing the objection that observed area differences in mobility may be driven by sorting (26).

Our first question is whether meaningful variation in equality and opportunity exists across areas populated by different ancestral groups. Fig. 1 A–E plots the distribution of ancestries across CZs. These maps reveal considerable clustering: Germans and Scandinavians are overrepresented in the Midwest, and Italian and French descendants in the Northeast, while those with British ancestry are scattered throughout the country. Present-day geographical clusters of European ancestral groups broadly reflect the historical patterns of European settlement during the Age of Mass Migration (SI Appendix, Fig. S2), which suggests a limited sorting over the 20th century.

Fig. 1 F–H displays measures of inequality and mobility (analogous maps for upward mobility among blacks and non-Hisp. whites, rags-to-riches mobility, as well as causal place effects, are displayed in SI Appendix, Fig. S3). Areas with a concentration of German and Scandinavian ancestry belong to the most upwardly mobile, while persistence is higher in areas with British ancestry. Moreover, CZs with higher rates of upward mobility, or lower rates of intergenerational persistence, tend to have a more equal distribution of incomes, in line with the GGC (4).

We assess the link between European origins and intergenerational mobility more directly in Fig. 2, where we also take race differences into account. Doing so is not only vital to accurately reflect the US opportunity structure (8), but also speaks directly to the hypothesis that mobility differences may extend beyond European descendants to local populations more broadly.

Fig. 2 displays binned scatter plots of the link between upward mobility and the concentration of each ancestry, distinguishing between outcomes for non-Hisp. whites and blacks. These correlations range from strongly positive for Scandinavian (non-Hisp. whites: $r=0.638$, 95% CI: $0.592,0.679$; blacks: $r=0.356$, 95% CI: $0.285,0.423$; SI Appendix, Table S2), to sharply negative for British ancestry shares (non-Hisp. whites: $r=-0.535$, 95% CI: $-0.585,-0.481$; blacks: $r=-0.132$, 95% CI: $-0.209,-0.053$; SI Appendix, Table S2). Several of these correlations are similar in size to important mobility correlates reported by Chetty et al. (4), including income inequality ($r=-0.58$), the fraction of black residents ($r=-0.58$), or teenage birth rates ($r=-0.61$).

We document cross-CZ gradients for the full range of mobility metrics in SI Appendix (Figs. S5 and S6 and the first column of Tables S8–S10). The correlations are similar, including for causal place effects—and, here, gradients are especially marked for children who grew up in the bottom half of the income distribution (SI Appendix, Fig. S5). Moreover, given that ancestry shares are heavily skewed (SI Appendix, Table S3), we confirm that patterns are similar when we transform each ancestry share to the ln of 1 plus the percentage who report a given ancestry (SI Appendix, Fig. S4).

Next, to allow a more direct comparison with the cross-national literature, we construct synthetic countries by aggregating economic outcomes across CZs weighted by their ancestry shares (Materials and Methods). For each European country, we thus obtain a synthetic US counterpart consisting of a weighted average of CZs.

Materials necessary to reproduce all findings can be accessed via a repository at the Open Science Framework (https://osf.io/5w7kf/) (29).

To estimate the ancestral composition of CZs, we use data from the 5% sample of the 1980 US Census made available through IPUMS (30), which align with the mobility metrics that are based on cohorts born in the early 1980s. Ancestry is defined by the Census Bureau as “a person’s ethnic origin, heritage, descent, or ‘roots,’ which may reflect their place of birth, place of birth of parents or ancestors, and ethnic identities that have evolved within the US” (31) and is based on self-identification. When defining European ancestry shares, we include all available subcategories (SI Appendix, section 1.1). To map respondents’ county group to CZs, we use cross-walks created by David Dorn (32).

The literature on economic mobility has generated a variety of measures, of which the most common is the intergenerational elasticity (IGE) of incomes. This statistic, which simply reflects the derivative of expected log child income ($\ln y_t$) with respect to log parent income ($\ln y_{t-1}$), is usually estimated by using ordinary least squares, yielding the expression:In Fig. 4, we use IGE estimates for sons born in ∼1960 and their fathers, derived from nationally representative data and corrected for measurement error, and estimates of income inequality from ∼1985, as reported by Corak (6). However, sensitivity to marginal distributions and the ages at which income is measured has led recent research to prefer the rank-order correlation (4), which represents a similar derivative where each variable has instead been transformed to percentile ranks, estimated as:In Figs. 1 and 4, we use data from Chetty et al. (4) on the rank–rank regression slope for US CZs, where children born in 1980–1982 and their parents are ranked in the national income distribution. Child income is measured as mean family income in 2011–2012 when children are approximately 30 y old, while parent income is measured by mean family income between 1996 and 2000. IGEs and rank correlations are similar in that they measure persistence of income throughout the distribution, but are uninformative about whether mobility is driven by improved prospects for those at the bottom. In Figs. 1 and 5, we therefore also use CZ-level data from Chetty et al. (4) on upward mobility for the 1980–1982 birth cohorts. Upward mobility is defined as the income-rank expectation for a child born to parents at the 25th percentile of the national income distribution, which equals the average rank of children born into the bottom half of the distribution given the linearity of the rank–rank relationship. In Figs. 2, 3, and 5, we use CZ-level measures of upward mobility separately by race for the 1978–1983 birth cohorts from Chetty et al. (8). In Fig. 5, we also present a conceptually similar CZ-level measure of upward mobility from Chetty et al. (4) for the 1980–1985 birth cohorts, which corresponds to the probability that a child born to parents in the bottom quintile of the national income distribution reaches the top quintile in adulthood. In addition, we use estimates of causal place effects from Chetty and Hendren (26) in Fig. 5. For each CZ, they estimate a causal exposure effect for children born to parents at the 25th (and 75th) income percentile, which corresponds to the expected percentage increase in household income at age 26 relative to the national mean from spending one additional year of childhood in that CZ. Note that all mobility measures are not available for the universe of CZs. Throughout the analysis, we always include all CZs in the contiguous United States with nonmissing data for each respective metric (n = 610–722; SI Appendix, Table S3).

In Figs. 1 and 4, we use CZ-level measures of income inequality obtained from Chetty et al. (4). Inequality is measured by the Gini coefficient of income among parents to children in the 1980–1982 birth cohorts, where incomes are top-coded at $100 million. Chetty et al. (4) report two versions of the CZ-level Gini coefficient, one calculated on the whole population and one subtracting the top 1% income share. We use the latter to reduce the influence of outliers and measurement error in top incomes. However, in SI Appendix, Fig. S8, we replicate the relationship using Gini coefficients based on the entire income distribution.

In Fig. 3, we display predicted state-level slopes from a mixed regression model of the form:where $\theta$ represents the CZ-level parameter of interest, modeled as a function of each CZ’s proportional representation of a given ancestry $a$ (subscript $c$ is for CZ). We let intercept (${\alpha }_s$) and slope (${\beta }_s$) coefficients vary at the state level (subscript $s$) and estimate models for each ancestry, as well as for black and non-Hisp. white mobility, separately (SI Appendix, section 2.5).

In Figs. 4 and 5, we present inequality and intergenerational mobility estimates for synthetic countries. When constructing synthetic countries, we weight the relevant population parameters at the CZ level as a function of each group’s proportional representation:where $\theta$ again represents the parameter of interest (subscript $a$ denotes ancestry), $a_c$ is the proportion of inhabitants who belong to a given ancestral group $a$, and $A_c$ is the proportion who report any of the ancestries in our study (SI Appendix, section 1.5). The bracketed expression places a lower weight on areas where other ancestries used in the analyses are heavily represented; alternative estimates excluding this penalty are reported in SI Appendix, section 2.8.