Intergenerational Income Mobility

Definition

Intergenerational income mobility is the degree to which a child's adult income is determined by their parents' income. High mobility means that parental income is a weak predictor of child income — children born poor can reach the top, and children born rich can fall. Low mobility means that the income distribution is reproduced across generations. In the United States, mobility is moderate by international standards at the national level but varies enormously by geography: some commuting zones have Scandinavian-level mobility while others have near-caste-like persistence.

Key Ideas

Rank-rank slope is the preferred mobility measure: slope of child income rank on parent income rank in the national distribution. Nationally: $0.341$ . Interpretation: a child whose parents move from the 10th to the 60th income percentile ( $50$ -pp improvement) is expected to have an income rank $\approx 17$ pp higher.
Intergenerational income elasticity (IGE) is the traditional measure: $\beta$ in $\log(Y_\text{child}) = \alpha + \beta \cdot \log(Y_\text{parent})$ . It is sensitive to (1) attenuation bias from short earnings windows that conflate transitory first-order autoregressive (AR(1)) shocks with permanent income, (2) life-cycle bias because the log-earnings proxy for permanent income is strongly age-dependent (Haider-Solon 2006: $\lambda_a \approx 0.2$ at age 25, peaks $\approx 1.0$ at 30–45, falls to $\approx 0.6$ by late 50s), and (3) classical measurement error in self-reported father earnings. After correcting via long administrative earnings averages, Mazumder (2005) finds US IGE rises from $\approx 0.25$ ( $T=2$ years) to $\approx 0.61$ ( $T=16$ years) — nearly double the 1990s consensus of $\approx 0.4$ . The rank-rank slope is preferred for cross-area or cross-time comparisons because it is bounded $[0,1]$ and invariant to earnings inequality differences across generations (Black and Devereux 2011).
Absolute upward mobility ( $\bar{r}_{25}$ ): Mean income rank of children whose parents were at the 25th percentile. This is the primary policy-relevant measure: it directly captures the average outcome of children starting near the bottom.
Relative mobility: The rank-rank slope estimated within a commuting zone. Measures how much of the local income distribution is reproduced across generations, regardless of the overall level.
U.S. absolute upward mobility is lower than peer countries: $P(Q_5|Q_1) = 7.5\%$ in the U.S. vs. $11.7\%$ in Denmark and $13.4\%$ in Canada.

How It Works

Measurement: Rank-Rank Slope

The rank-rank slope $\beta$ estimates the best linear predictor of child income rank given parent income rank. Because it uses ranks (not dollars), it is:

Invariant to the treatment of zero-income households
Comparable across areas with different income distributions
Bounded: a slope of $0$ means perfect mobility; a slope of $1$ means perfect immobility

At $0.341$ nationally, the U.S. rank-rank slope means that roughly one-third of the income-rank position of parents is transmitted to children on average.

Geographic Variation in Absolute Upward Mobility

Commuting Zone (CZ), 50 largest	$\bar{r}_{25}$
Salt Lake City, UT	46.2
San Jose, CA	45.1
Washington, DC	45.1
Seattle, WA	44.5
Minneapolis, MN	44.5
New York, NY	43.0
Los Angeles, CA	42.0
Indianapolis, IN	40.9
Columbus, OH	40.5
Raleigh, NC	40.1
Atlanta, GA	36.4
Charlotte, NC	35.8

The $10.4$ percentile-rank gap between Salt Lake City and Charlotte is enormous: a child born at the 25th percentile in Charlotte has expected adult income at the $35.8$ th percentile; the same child born in Salt Lake City reaches the $46.2$ nd percentile. Both are raised at the same family income level; geography accounts for the difference.

Geographic Pattern

The Southeast and Appalachian/Ohio Valley corridor have the lowest absolute mobility. The Great Plains, Mountain West, and rural upper Midwest have the highest. The pattern is not fully explained by race, poverty rates, or urbanization — after controlling for observable characteristics, meaningful geographic residuals remain. This implies that local institutions, norms, and physical structure (rather than just population composition) shape mobility outcomes.

Five Correlates of Upward Mobility

Correlate	$r$ with $\bar{r}_{25}$	Direction
Fraction single parents	$-0.76$	Highest single-parent share, lowest mobility
Social capital index	$0.641$	More civic engagement, higher mobility
School quality (output)	$\approx 0.60$	Better test scores/lower dropout, higher mobility
Bottom-99 Gini	$-0.634$	More income inequality among non-rich, lower mobility
Residential segregation	$-0.605$	More income segregation, lower mobility
Top 1% income share	$-0.19$	Nearly uncorrelated

The near-zero correlation with the top 1% income share is striking: it is the distribution of income among the broad population, not concentration at the apex, that predicts whether children escape poverty.

Demographic Backdrop: Wu (2008) Cohort Trends

The fraction-single-parents correlate reflects a 50-year structural shift, not a short-run artifact. Wu (2008) documents that the share of U.S. women with at least one nonmarital birth by age 30 rose monotonically from $\approx 10\%$ (1925–1929 cohort) to $\approx 27\%$ (1965–1969 cohort), with white women rising from $\approx 7\%$ to $\approx 20\%$ . This long-run cohort trend is the demographic driver underlying the cross-CZ variation that Chetty et al. measure. The CZs with the lowest upward mobility — the Southeast, Appalachian corridor — are also the CZs where nonmarital fertility has been highest for longest. See Nonmarital Fertility.

Community-Level Family Structure Effect

The strongest correlate — fraction of single parents — operates at the community level, not just the individual level. Children of two-parent households living in high-single-parent CZs have lower expected mobility than children of two-parent households living in low-single-parent CZs. This implies a peer, network, or norm channel: growing up surrounded by non-married households reduces a child's upward mobility even if their own parents are married. This is a social multiplier effect that individual-level family policy cannot fully address.

The reverse also holds: the Black–white mobility gap is partially a CZ-composition effect. Whites living in high-Black-share CZs have lower mobility than whites in low-Black-share CZs. The racial composition of the local area predicts mobility for all races — suggesting that the conditions correlated with high Black share (concentrated poverty, segregation, underfunded schools) are the proximate mechanism.

Cross-Country IGE (Jäntti et al. 2006)

Country	IGE
Denmark	0.071
Norway	0.155
Finland	0.173
Sweden	0.258
United Kingdom	0.306
United States	0.517

The Nordic $<$ UK $<$ US ordering is robust across samples and methods. The US IGE of $\approx 0.52$ (rising to $\approx 0.61$ after Mazumder correction) is roughly $7\times$ Denmark's, confirming that the US is exceptional in its low intergenerational mobility relative to peer high-income countries, not its high mobility as political mythology suggests.

Causal Identification of Intergenerational Transmission

The identification hierarchy for separating causal environment effects from genetic/selection confounding runs from instrumental variables (IV) (compulsory schooling laws), through adoptees, through within-twin family fixed effects, to ordinary least squares (OLS). All causal estimates fall below OLS.

Education transmission: Black, Devereux, and Salvanes (2005) use Norwegian compulsory-schooling reform as an IV. OLS elasticity $\approx 0.22$ ; IV $\approx 0.04$ and statistically indistinguishable from zero. Most of the OLS association reflects genetic or selection confounding, not causal human capital transmission.

Random adoptee evidence: Sacerdote (2007) studies Korean-American children placed with random adoptive families. Mother's education coefficient $= 0.32$ for biological children vs. $0.09$ for adoptees; father's income coefficient $= 0.09$ biological vs. $\approx 0.00$ for adoptees. Environmental nurture effects on education are small; genetics/selection accounts for the overwhelming majority of OLS associations.

Structural twin decomposition (Björklund et al. 2005): Genetic variance share $g^2 = 0.28$ , shared-family-environment share $s^2 = 0.04$ for brothers' earnings. Non-shared environment accounts for the remaining $\approx 0.68$ . The shared family environment that IGE ostensibly captures is empirically small; most of the remaining variance is non-shared environment (luck, idiosyncratic schooling, peer effects) and genetics.

Causal Neighborhood Effects: The Chetty-Hendren Movers Design

Chetty and Hendren (2018a/b) resolve the long-standing debate between observational studies (which found large neighborhood effects) and experimental studies like Moving to Opportunity (MTO) (which found small effects). Using 7+ million U.S. families who moved across commuting zones, they exploit variation in the age at which children move: siblings who move to the same destination at different ages receive different amounts of neighborhood exposure, and their adult income outcomes diverge in proportion to the age gap × destination quality at a rate of $\approx$ $4\%$ per year of childhood exposure. Moving at birth to a neighborhood where permanent residents score 1 percentile higher → child captures $\approx 80\%$ of that advantage over 20 years; there is no sharp critical age, so improvements matter at all stages of childhood. Three placebo tests (cohort-specific, gender-specific, and quantile-specific convergence) confirm the causal interpretation — the pattern is too multi-dimensional to arise from selection. The reconciliation with MTO: MTO families moved when children were already $\approx$ age 8, leaving only $\approx 12$ years of exposure → $\approx 48\%$ pickup, not a null effect. See Chetty and Hendren 2018 — The Effects of Neighborhoods on Intergenerational Mobility.

Causality Warning (Correlates)

These are correlates, not causal estimates. Family structure might itself be downstream of economic conditions (manufacturing decline → male idleness → non-marriage, per Marriage Market and Autor, Dorn, and Hanson (ADH) 2019). Residential segregation might reflect historical housing policy (redlining, exclusionary zoning) that independently shaped both mobility and social capital. The paper is explicit: it provides a map of where mobility is high and low and what covaries with it, but it does not identify why through causal mechanisms.

Family Disadvantage, Gender, and Educational Pathways (Autor et al. 2019)

Autor, Figlio, Karbownik, Roth, and Wasserman (2019) provide a partial causal decomposition of the family structure → mobility link, focusing on the gender-specific channel. Using linked Florida administrative data, they show that boys in low-socioeconomic status (SES) (and single-parent) families have substantially worse behavioral and educational outcomes than their sisters — a gap that is confirmed to be postnatal and to operate largely within rather than between schools and neighborhoods. Each standard deviation (SD) increase in the SES index closes the boy-girl graduation gap by $1.5$ pp, and behavioral outcomes (suspensions) account for $\approx 90.6\%$ of the mediator path from family disadvantage to dropout. Crucially, school and neighborhood controls reduce the family SES effect by $\leq 25\%$ , establishing that family environment is a distinct input to child outcomes independent of the Chetty-Hendren neighborhood channel. The implication for intergenerational mobility is that the fraction-single-parents correlate identified by Chetty et al. (2014) operates partly through a direct behavioral-pathway effect on boys' educational attainment, not purely through school-quality or neighborhood composition effects. See Gender Gap in Educational Outcomes.

Why It Matters

The U.S. as a Collection of Societies

The paper's central policy reframing: talking about "U.S. intergenerational mobility" as a single number is misleading. Some CZs have European-level mobility; others rival emerging-market persistence. Policy interventions appropriate for Charlotte are not the same as those appropriate for Salt Lake City — and treating them identically obscures this heterogeneity.

Geographic Connection to the Income-Mortality Gradient

The low-mobility Southeast/Appalachian belt documented here is the same geography documented in Income-Mortality Gradient (Chetty et al. 2016) and Deaths of Despair (Case and Deaton 2015/2017) as the region with the worst health outcomes for low-income adults. The shared geography suggests common upstream drivers: manufacturing decline, low educational attainment, religious and social capital erosion, concentrated poverty. The children who grow up immobile in the Southeast are the same adults who will appear in the low-income mortality statistics two or three decades later.

Implications for Inequality Policy

The finding that the top 1% income share is nearly uncorrelated with mobility ( $r = -0.19$ ) while broad inequality (bottom-99 Gini) is strongly correlated ( $r = -0.634$ ) has direct policy implications. Policies targeting the very top of the income distribution (top marginal rates, wealth taxes) are not obviously the levers that affect intergenerational mobility at the bottom. The distributional action is in the bottom and middle of the income distribution — school quality, residential segregation, and the social conditions of low-income communities.

Family Structure and the Limits of Individual-Level Policy

The community-level family structure effect implies that policies targeting individual families (marriage promotion programs, parenting interventions) face a structural constraint: the surrounding community matters independently of what any one family does. Area-level interventions — reducing residential segregation, improving local school quality, rebuilding social institutions — may be necessary complements to individual-level support.

Open Questions

What is the causal mechanism behind the community-level family structure effect? Is it peer networks (children model non-married adults), school composition (behavioral spillovers from children of single parents), or labor market networks (single parents transmit fewer job connections)?
Has geographic mobility variation changed since the 1980–1991 birth cohorts? The Opportunity Atlas (Chetty et al. 2018) provides some updates, but the time-series of CZ-level mobility trends remains an active research area.
Does improving local labor market conditions (reversing manufacturing decline) increase mobility within a generation, or are neighborhood effects slow-moving?
The paper captures children's outcomes at age 28–32. What is the stability of the geographic ranking at later ages (35–45), when earnings are more settled?