Social capital I: measurement and associations with economic mobility

The study investigates which forms of social capital are most relevant for economic opportunity, focusing on the role of connections across socioeconomic lines. Prior research emphasizes social capital as a potential driver of outcomes including education, health, and intergenerational mobility, but has been limited by small-scale or indirect measures of social networks. The authors leverage large-scale Facebook friendship data to construct new community-level measures of social capital in the United States and assess their associations with upward income mobility. The central research question is whether economic connectedness—friendships between low- and high-SES individuals—better predicts upward mobility than other concepts of social capital such as network cohesiveness or civic engagement. The purpose is to provide granular, validated measures of social capital, document their geographic variation, and evaluate their relationships with intergenerational mobility to inform policy and future research.

The paper builds on theoretical and empirical work that posits benefits of social ties to well-resourced individuals for information flow, aspiration formation, mentorship, and job referrals. Classic and contemporary theories distinguish bridging versus bonding social capital and link network structure to outcomes such as norm enforcement, learning, and diffusion. Empirical evidence includes peer effects, labor-market impacts of networks, and the role of network diversity in development. Prior datasets like Add Health provided valuable but limited network snapshots, and mobile phone-based measures captured experienced segregation without directly observing friendship ties. Work on intergenerational mobility has linked mobility differences to segregation, poverty, and inequality, but lacked direct measures of cross-SES interaction. This study addresses these gaps by using large-scale friendship data and constructing multiple social capital indices to disentangle their relationships with economic mobility.

Data source and sample: The authors use privacy-protected Facebook data on the social networks of 72.2 million U.S.-based users aged 25–44, active within the last 30 days, with at least 100 U.S.-based friends, and a non-missing residential ZIP code. Representativeness is assessed against national surveys; users are used as proxies for real-world friendships, not online interaction effects. Locations (ZIP code, county), users' SES, and parents' SES are derived and described in the Methods. SES construction: SES is computed via a machine-learning index combining information from 22 variables (including neighborhood median income and self-reported education), then ranking individuals within birth cohorts by national SES percentile. Validation shows high correlation with external income measures; simpler SES proxies yield similar results. Social capital measures: Three categories are constructed. (1) Cross-type connectedness: primary measure is economic connectedness (EC), defined as twice the share of above-median-SES friends among below-median-SES individuals in a community. Analogous measures are built for high-SES individuals and by finer SES deciles; additional connectedness by language and age is computed. (2) Network cohesiveness: clustering (average share of friend pairs who are also friends), support ratio (share of friendships with at least one mutual friend), and spectral homophily (second eigenvalue of the row-stochasticized adjacency matrix, capturing fragmentation). (3) Civic engagement: volunteering rate (share of users who are members of volunteering or activism groups on Facebook) and density of civic organizations (Facebook pages for civic organizations per 1,000 users). Geographic aggregation: Statistics are constructed for counties and ZIP codes (ZIP code tabulation areas), with reliability assessed; maps and public release at socialcapital.org. Childhood EC: Measured using high school friendships and parental SES for Facebook users, and separately using Instagram data for current 13–17 year olds (parental SES proxied by ZIP code and phone model). Outcome measures: Upward income mobility is drawn from the Opportunity Atlas, defined as the mean adult income percentile at age ~35 for children with parents at the 25th percentile of national income (1978–1983 cohorts). Additional outcomes include high school graduation and teen birth rates. Empirical approach: Univariate and multivariable correlations and OLS regressions of upward mobility on social capital measures are estimated at county and ZIP levels, weighted by the number of children with below-median parental income; standard errors are clustered by commuting zone. Nonparametric relationships are visualized via scatter/binned plots. Robustness and mechanisms: Reverse causality is addressed by correlating mobility with childhood EC (pre-labor-market friendships). Selection is probed by restricting to racially homogeneous areas and by relating EC to quasi-experimental county-level causal effects on mobility from Chetty and Hendren (based on age-at-move identification). Controls include median income, poverty rates, racial and income segregation indices, income inequality (Gini excluding top 1% share), school test scores, job availability/growth, and family structure. Incremental explanatory power is assessed via incremental R-squared, Lasso selection, and within-county ZIP code analyses.

• Strong SES homophily: A one-percentile increase in an individual's SES rank is associated with a 0.44 percentile increase in the mean SES rank of friends (0.46 using the top 10 closest friends). Among high school friendships, homophily by parental SES has a slope of 0.31 in both Facebook and Add Health data. • Composition of friendships: On average, 38.8% of friends of below-median-SES individuals have above-median SES, versus 70.6% for above-median-SES individuals; relative under-representation of high-SES friends among low-SES individuals is 22.4%, and over-representation among high-SES individuals is 41.2%. Benchmarking to equal friend counts implies low-SES individuals make about 30.2% fewer high-SES friends than expected absent homophily. Bottom-decile users under-represent top-decile friends by roughly 75% relative to population share. • Geographic variation in EC: County EC ranges from below 0.58 (bottom decile) to at least 1.05 (top decile); EC is generally lowest in the Southeast/Southwest and certain Midwestern cities, and highest in the rural Midwest and East Coast. Within-county ZIP code variation is substantial (42% of ZIP-level EC variation is within counties); Los Angeles ZIP codes range from 0.62 to 1.25 between the 10th and 90th percentiles. • Distinct dimensions: Connectedness by language and age and cohesiveness measures exhibit different spatial patterns; across-county correlations with EC are low (e.g., language connectedness correlates 0.10 with EC). • Civic engagement: Volunteering rates and civic organization density vary markedly and correlate moderately with EC (volunteering-EC correlation 0.46; civic organizations-Penn State index correlation 0.67), but are weaker predictors of mobility. • EC and upward mobility: EC correlates strongly with upward mobility across counties (correlation 0.65, s.e. 0.04) and across ZIP codes (0.69). In the 200 most populous counties, the OLS slope implies that a 1.0 increase in EC is associated with a 16.4 percentile increase in adult income rank; a 0.5 EC increase corresponds to about an 8.2 percentile gain. Moving from the 10th to 90th percentile ZIP code in EC is associated with an 11 percentile increase in adult income rank for children from low-income families. • Childhood EC and reverse causality: Correlations between upward mobility and childhood EC remain high (0.44 using Facebook parental SES; 0.62 using Instagram-based measure), indicating adult EC-mobility links are not driven mainly by reverse causality. • Quasi-experimental evidence: County-level causal mobility effects (from age-at-move research) correlate 0.44 (s.e. 0.06) with EC. A 1.0 unit higher EC is estimated to increase adult income by 9.8 percentiles for children of low-income parents (about 30.7% relative to mean ranks). Increasing EC by 0.57 (10th to 90th percentile ZIP) predicts a 17.5% income increase; aligning low-SES children's EC with high-SES children's average EC implies about a 19.5% income increase, closing roughly 37% of the 17-percentile gap between children of parents at the 25th and 75th percentiles. • Mediation and controls: EC remains a strong predictor of mobility when controlling for median income or poverty rates; those income measures lose much of their predictive power. EC also attenuates or eliminates associations of mobility with racial composition, racial and income segregation, and income inequality (Great Gatsby curve). Differences in EC explain much of the negative association between the share of Black residents or higher inequality and mobility. • Heterogeneity by SES of children: Greater cross-SES connectedness strongly benefits children from low-income families; conditional on income mix, it does not harm outcomes for children from high-income families, suggesting benefits can accrue without adverse effects on higher-SES children. • Network cohesiveness and civic engagement: These measures are not robustly associated with mobility across areas, and their relationships with mobility vary by context; by contrast, EC-mobility relationships are consistently positive across nearly all counties.

The findings directly address the central question: among diverse forms of social capital, economic connectedness—cross-SES friendships—is most strongly and consistently associated with upward income mobility. This relationship persists across geographic scales and is robust to controls for neighborhood income levels, segregation, inequality, school quality proxies, labor market indicators, and family structure. Analyses of childhood EC mitigate reverse causality concerns, and quasi-experimental county effect estimates indicate that growing up in higher-EC places causally improves adult economic outcomes for low-income children. EC appears to mediate the relationships between mobility and other neighborhood characteristics: lower neighborhood income, higher racial or income segregation, and greater inequality are linked to lower mobility largely insofar as they correlate with reduced cross-SES interaction. The stability of EC-mobility associations within counties and across ZIP codes underscores its relevance for place-based policy. While network cohesiveness and civic engagement capture important social features and predict other outcomes (for example, health and life expectancy), they do not explain spatial differences in intergenerational mobility as strongly as EC. The results suggest that policies and community efforts that increase cross-class interaction could improve economic opportunity for disadvantaged children without imposing significant costs on higher-SES peers when income composition is held constant.

This paper constructs and publicly releases granular, privacy-preserving measures of social capital across U.S. communities, distinguishing economic connectedness, network cohesiveness, and civic engagement. The measures reveal substantial geographic variation and low cross-measure correlations, demonstrating that social capital is multi-dimensional. The key contribution is showing that economic connectedness is a powerful and robust predictor of upward income mobility, and that its effects are likely causal based on childhood exposure and quasi-experimental evidence. EC also helps reconcile prior findings on the roles of poverty, segregation, and inequality by mediating their relationships with mobility. Future research directions include: (1) assessing which forms of social capital matter for other outcomes such as health behaviors and political preferences; (2) extending these measurement methods to other countries and alternative data sources; and (3) directly testing interventions to increase economic connectedness to improve intergenerational mobility. The companion paper examines determinants of EC and potential policy levers to raise EC among low-SES individuals.

• Data source and representativeness: Measures rely on Facebook friendship data for users aged 25–44; while benchmarking suggests reasonable representativeness, certain populations are underrepresented, and results reflect this user base. • Measurement of SES: Individual SES is a proxy index derived from multiple signals rather than direct income; although validated, measurement error may remain. • Network proxy: Friendships on Facebook are used as proxies for real-world ties; effects of online networks per se are not studied. • Ecological focus: Analyses are primarily area-level; individual-level regressions using Facebook income proxies may suffer greater measurement error. • Causality: Although childhood EC correlations and quasi-experimental county effects support causal interpretations, unobserved confounding at finer scales cannot be fully ruled out. • Exposure constraints: In very low-income areas with few high-SES residents, scope for increasing EC via local ties may be mechanically limited by exposure, necessitating non-local strategies.